U.S. patent application number 13/800155 was filed with the patent office on 2013-10-17 for level decision in rate distortion optimized quantization.
This patent application is currently assigned to QUALCOMM INCORPORATED. The applicant listed for this patent is QUALCOMM INCORPORATED. Invention is credited to Muhammed Zeyd Coban, Marta Karczewicz, Yang Yu.
Application Number | 20130272385 13/800155 |
Document ID | / |
Family ID | 49325054 |
Filed Date | 2013-10-17 |
United States Patent
Application |
20130272385 |
Kind Code |
A1 |
Yu; Yang ; et al. |
October 17, 2013 |
LEVEL DECISION IN RATE DISTORTION OPTIMIZED QUANTIZATION
Abstract
A computing device, such as a video encoder, determines an
initial quantized level for a coefficient of a coefficient block
and determines whether the coefficient is less than the product of
the initial quantized level and a quantization step size value. In
response to determining that the coefficient is less than the
product of the initial quantized level and the quantization step
size value, the computing device determines rate-distortion costs
of quantizing the coefficient to be the initial quantized level for
the coefficient, the initial quantized level minus one, and in some
circumstances, 0. The computing device determines an actual
quantized level for the coefficient based at least in part on the
calculated rate-distortion costs and includes the actual quantized
level in a quantized version of the coefficient block.
Inventors: |
Yu; Yang; (San Diego,
CA) ; Coban; Muhammed Zeyd; (Carlsbad, CA) ;
Karczewicz; Marta; (San Diego, CA) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
QUALCOMM INCORPORATED |
San Diego |
CA |
US |
|
|
Assignee: |
QUALCOMM INCORPORATED
San Diego
CA
|
Family ID: |
49325054 |
Appl. No.: |
13/800155 |
Filed: |
March 13, 2013 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
61623929 |
Apr 13, 2012 |
|
|
|
61623948 |
Apr 13, 2012 |
|
|
|
61624116 |
Apr 13, 2012 |
|
|
|
Current U.S.
Class: |
375/240.03 |
Current CPC
Class: |
H04N 19/149 20141101;
H04N 19/172 20141101; H04N 19/126 20141101; H04N 19/18 20141101;
H04N 19/147 20141101 |
Class at
Publication: |
375/240.03 |
International
Class: |
H04N 7/26 20060101
H04N007/26 |
Claims
1. A method of encoding video data, the method comprising:
determining an initial quantized level for a coefficient of a
coefficient block; determining whether the coefficient is less than
the product of the initial quantized level and a quantization step
size value; and in response to determining that the coefficient is
less than the product of the initial quantized level and the
quantization step size value: determining rate-distortion costs of
quantizing the coefficient to be the initial quantized level and
the initial quantized level minus one; and determining an actual
quantized level for the coefficient based at least in part on the
calculated rate-distortion costs.
2. The method of claim 1, further comprising determining, without
determining the rate-distortion costs and in response to
determining that the coefficient is not less than the product of
the initial quantized level and the quantization step size value,
that the actual quantized level for the coefficient is the initial
quantized level for the coefficient.
3. The method of claim 1, further comprising: in response to
determining that the coefficient is less than the product of the
initial quantized level and the quantization step size value:
determining, in response to determining that the initial quantized
level is less than three, the rate-distortion costs of quantizing
the coefficient to be the initial quantized level for the
coefficient, the initial quantized level minus one, and a
rate-distortion cost of quantizing the coefficient to be zero; and
determining, in response to determining that the initial quantized
level is greater than or equal to three, the rate-distortion costs
of quantizing the coefficient to be the initial quantized level and
the initial quantized level minus one.
4. The method of claim 1, wherein the coefficient occurs after a
first non-zero coefficient in a reverse scan order in the
coefficient block.
5. The method of claim 1, further comprising determining a last
significant coefficient, the last significant coefficient being the
first non-zero coefficient of the coefficient block in a reverse
scan order.
6. The method of claim 1, further comprising: entropy encoding the
actual quantized level; and outputting a bitstream that includes
the entropy-encoded actual quantized level.
7. The method of claim 1, wherein determining the actual quantized
level comprises selecting an actual quantized level that produces a
lowest rate-distortion cost.
8. The method of claim 7, further comprising generating a
reconstructed value of the coefficient by multiplying the actual
quantized level by the quantization step size value.
9. The method of claim 1, wherein determining the initial quantized
level for the coefficient comprises calculating z = w + f .DELTA.
sign ( W ) ##EQU00006## where z is the initial quantized level for
the coefficient, W is an initial value of the coefficient, f is a
quantization offset for the coefficient, .DELTA. is a quantization
step, [] denotes rounding to the nearest integer towards minus
infinity and sign( ) is a function that returns the sign of the
initial value of the coefficient.
10. The method of claim 1, wherein: the method further comprises:
determining, based at least in part on entropies of context models
of a previous frame, rate values for the context models; and
generating a table that indicates, for each of the context models,
a value of .lamda. multiplied by the rate value for the context
model, where .lamda. is a fixed value; and wherein determining the
rate-distortion costs comprises determining, based at least in part
on the values in the table, the rate-distortion costs of quantizing
the coefficient to be the initial quantized level for the
coefficient, the initial quantized level minus one, and zero.
11. A computing device that encodes video data, the computing
device comprising one or more processors that are configured to:
determine an initial quantized level for a coefficient of a
coefficient block; determine whether the coefficient is less than
the product of the initial quantized level and a quantization step
size value; and in response to determining that the coefficient is
less than the product of the initial quantized level and the
quantization step size value: determine rate-distortion costs of
quantizing the coefficient to be the initial quantized level and
the initial quantized level minus one; and determine an actual
quantized level for the coefficient based on the calculated
rate-distortion costs.
12. The computing device of claim 11, wherein the one or more
processors are further configured to determine, without determining
the rate-distortion costs and in response to determining that the
coefficient is not less than the product of the initial quantized
level and the quantization step size value, that the actual
quantized level for the coefficient is the initial quantized level
for the coefficient.
13. The computing device of claim 11, wherein the one or more
processors are configured to: in response to determining that the
coefficient is less than the product of the initial quantized level
and the quantization step size value: determine, in response to
determining that the initial quantized level is less than three,
the rate-distortion costs of quantizing the coefficient to be the
initial quantized level for the coefficient, the initial quantized
level minus one, and a rate-distortion cost of quantizing the
coefficient to be zero; and determine, in response to determining
that the initial quantized level is greater than or equal to three,
the rate-distortion costs of quantizing the coefficient to be the
initial quantized level and the initial quantized level minus
one.
14. The computing device of claim 11, wherein the coefficient
occurs after a first non-zero coefficient in a reverse scan order
in the coefficient block.
15. The computing device of claim 11, wherein the one or more
processors are further configured to determine a last significant
coefficient, the last significant coefficient being the first
non-zero coefficient in a reverse scan order.
16. The computing device of claim 11, wherein the one or more
processors are further configured to: entropy encode the actual
quantized level; and output a bitstream that includes the
entropy-encoded actual quantized level.
17. The computing device of claim 11, wherein the one or more
processors are configured to determine the actual quantized level
at least in part by selecting an actual quantized level that
produces a lowest rate-distortion cost.
18. The computing device of claim 17, wherein the one or more
processors are further configured to generate a reconstructed value
of the coefficient by multiplying the actual quantized level by the
quantization step size value.
19. The computing device of claim 11, wherein the one or more
processors are configured to determine the initial quantized level
for the coefficient by calculating z = w + f .DELTA. sign ( W )
##EQU00007## where z is the initial quantized level for the
coefficient, W is an initial value of the coefficient, f is a
quantization offset for the coefficient, .DELTA. is a quantization
step, [] denotes rounding to the nearest integer towards minus
infinity and sign( ) is a function that returns the sign of the
initial value of the coefficient.
20. The computing device of claim 11, wherein the one or more
processors are configured to: determine, based at least in part on
entropies of context models of a previous frame, rate values for
the context models; generate a table that indicates, for each of
the context models, a value of .lamda. multiplied by the rate value
for the context model, where .lamda. is a fixed value; and
determine, based at least in part on the values in the table, the
rate-distortion costs of quantizing the coefficient to be the
initial quantized level for the coefficient, the initial quantized
level minus one, and zero.
21. A computing device comprising: means for determining an initial
quantized level for a coefficient of a coefficient block; means for
determining whether the coefficient is less than the product of the
initial quantized level and a quantization step size value; means
for determining, in response to determining that the coefficient is
less than the product of the initial quantized level and the
quantization step size value, rate-distortion costs of quantizing
the coefficient to be the initial quantized level and the initial
quantized level minus one; and means for determining, in response
to determining that the coefficient is less than the product of the
initial quantized level and the quantization step size value, an
actual quantized level for the coefficient based on the calculated
rate-distortion costs.
22. A computer-readable storage medium that stores instructions
that, when executed by one or more processors of a computing
device, configure the computing device to: determine an initial
quantized level for a coefficient of a coefficient block; determine
whether the coefficient is less than the product of the initial
quantized level and a quantization step size value; and in response
to determining that the coefficient is less than the product of the
initial quantized level and the quantization step size value:
determine rate-distortion costs of quantizing the coefficient to be
the initial quantized level and the initial quantized level minus
one; and determine an actual quantized level for the coefficient
based on the calculated rate-distortion costs.
Description
[0001] This application claims the benefit of U.S. Provisional
Patent Application No. 61/623,929, filed Apr. 13, 2012, U.S.
Provisional Patent Application No. 61/623,948, filed Apr. 13, 2012,
and U.S. Provisional Patent Application No. 61/624,116, filed Apr.
13, 2012, the entire content of each of which is incorporated
herein by reference.
TECHNICAL FIELD
[0002] This disclosure relates to video coding and compression and,
in particular, to the implementation of rate distortion optimized
quantization in video coding.
BACKGROUND
[0003] Digital video capabilities can be incorporated into a wide
range of devices, including digital televisions, digital direct
broadcast systems, wireless broadcast systems, personal digital
assistants (PDAs), laptop or desktop computers, tablet computers,
e-book readers, digital cameras, digital recording devices, digital
media players, video gaming devices, video game consoles, cellular
or satellite radio telephones, so-called "smart phones," video
teleconferencing devices, video streaming devices, and the like.
Digital video devices implement video compression techniques, such
as those described in the standards defined by MPEG-2, MPEG-4,
ITU-T H.263, ITU-T H.264/MPEG-4, Part 10, Advanced Video Coding
(AVC), the High Efficiency Video Coding (HEVC) standard presently
under development, and extensions of such standards. The video
devices may transmit, receive, encode, decode, and/or store digital
video information more efficiently by implementing such video
compression techniques.
[0004] Video compression techniques perform spatial (intra-picture)
prediction and/or temporal (inter-picture) prediction to reduce or
remove redundancy inherent in video sequences. For block-based
video coding, a video slice (i.e., a video frame or a portion of a
video frame) may be partitioned into video blocks, which may also
be referred to as tree blocks, coding units (CUs) and/or coding
nodes. Video blocks in an intra-coded (I) slice of a picture are
encoded using spatial prediction with respect to reference samples
in neighboring blocks in the same picture. Video blocks in an
inter-coded (P or B) slice of a picture may use spatial prediction
with respect to reference samples in neighboring blocks in the same
picture or temporal prediction with respect to reference samples in
other reference pictures. Pictures may be referred to as frames,
and reference pictures may be referred to a reference frames.
[0005] Spatial or temporal prediction results in a predictive block
for a block to be coded. Residual data represents pixel differences
between the original block to be coded and the predictive block. An
inter-coded block is encoded according to a motion vector that
points to a block of reference samples forming the predictive
block, and the residual data indicates the difference between the
coded block and the predictive block. An intra-coded block is
encoded according to an intra-coding mode and the residual data.
For further compression, the residual data may be transformed from
the pixel domain to a transform domain, resulting in residual
coefficients, which then may be quantized. The quantized
coefficients, initially arranged in a two-dimensional array, may be
scanned in order to produce a one-dimensional vector of
coefficients, and entropy coding may be applied to achieve even
more compression.
SUMMARY
[0006] In general, this disclosure describes quantization
techniques. In some examples, a computing device, such as a video
encoder, determines an initial quantized level for a coefficient of
a coefficient block and determines whether the coefficient is less
than the product of the initial quantized level and a quantization
step size value. In response to determining that the coefficient is
less than the product of the initial quantized level and the
quantization step size value, the computing device determines
rate-distortion costs of quantizing the coefficient to be the
initial quantized level for the coefficient, the initial quantized
level minus one, and in some circumstances, 0. The computing device
determines an actual quantized level for the coefficient based at
least in part on the calculated rate-distortion costs. The
computing device includes the actual quantized level in a quantized
version of the coefficient block.
[0007] In one example, this disclosure describes a method of
encoding video data. The method comprises determining an initial
quantized level for a coefficient of a coefficient block. The
method also comprises determining whether the coefficient is less
than the product of the initial quantized level and a quantization
step size value. Furthermore, the method comprises in response to
determining that the coefficient is less than the product of the
initial quantized level and the quantization step size value,
determining rate-distortion costs of quantizing the coefficient to
be the initial quantized level and the initial quantized level
minus one. In addition, the method comprises, in response to
determining that the coefficient is less than the product of the
initial quantized level and the quantization step size value,
determining an actual quantized level for the coefficient based at
least in part on the calculated rate-distortion costs.
[0008] In another example, this disclosure describes a computing
device that encodes video data. The computing device comprises one
or more processors that are configured to determine an initial
quantized level for a coefficient of a coefficient block. The one
or more processors are also configured to determine whether the
coefficient is less than the product of the initial quantized level
and a quantization step size value. The one or more processor are
configured such that in response to determining that the
coefficient is less than the product of the initial quantized level
and the quantization step size value, the one or more processors
determine rate-distortion costs of quantizing the coefficient to be
the initial quantized level and the initial quantized level minus
one. Furthermore, in response to determining that the coefficient
is less than the product of the initial quantized level and the
quantization step size value, the one or more processors determine
an actual quantized level for the coefficient based on the
calculated rate-distortion costs.
[0009] In another example, this disclosure describes a computing
device comprising means for determining an initial quantized level
for a coefficient of a coefficient block. In addition, the
computing device comprises means for determining whether the
coefficient is less than the product of the initial quantized level
and a quantization step size value. Furthermore, the computing
device comprises means for determining, in response to determining
that the coefficient is less than the product of the initial
quantized level and the quantization step size value,
rate-distortion costs of quantizing the coefficient to be the
initial quantized level and the initial quantized level minus one.
The computing device also comprises means for determining, in
response to determining that the coefficient is less than the
product of the initial quantized level and the quantization step
size value, an actual quantized level for the coefficient based on
the calculated rate-distortion costs.
[0010] In another example, this disclosure describes a
computer-readable storage medium that stores instructions that,
when executed by one or more processors of a computing device,
configure the computing device to determine an initial quantized
level for a coefficient of a coefficient block. The instructions
also configure the computing device to determine whether the
coefficient is less than the product of the initial quantized level
and a quantization step size value. The instructions also configure
the computing device such that in response to determining that the
coefficient is less than the product of the initial quantized level
and the quantization step size value, determines rate-distortion
costs of quantizing the coefficient to be the initial quantized
level and the initial quantized level minus one. Furthermore, the
instructions configure the computing device such that in response
to determining that the coefficient is less than the product of the
initial quantized level and the quantization step size value, the
computing device determines an actual quantized level for the
coefficient based on the calculated rate-distortion costs.
[0011] The details of one or more examples of the disclosure are
set forth in the accompanying drawings and the description below.
Other features, objects, and advantages will be apparent from the
description, drawings, and claims.
BRIEF DESCRIPTION OF DRAWINGS
[0012] FIG. 1 is a block diagram illustrating an example video
coding system that may utilize the techniques described in this
disclosure.
[0013] FIG. 2 is a block diagram illustrating an example video
encoder that may implement the techniques described in this
disclosure.
[0014] FIG. 3 is a block diagram illustrating an example video
decoder that may implement the techniques described in this
disclosure.
[0015] FIG. 4 is a conceptual diagram illustrating scalar
quantization with deadzone.
[0016] FIG. 5 is a conceptual diagram illustrating quantization
offset and level decision.
[0017] FIG. 6 is a flowchart illustrating an example operation to
quantize a coefficient, in accordance with one or more techniques
of this disclosure.
[0018] FIG. 7 is a flowchart illustrating an example operation to
quantize coefficients of a coefficient block, in accordance with
one or more techniques of this disclosure.
[0019] FIG. 8 is a flowchart illustrating an example portion of the
operation of FIG. 7.
[0020] FIG. 9 is a flowchart illustrating another example operation
to quantize coefficients of a coefficient block, in accordance with
one or more techniques of this disclosure.
[0021] FIG. 10 is a flowchart illustrating another example
operation to quantize coefficients of a coefficient block, in
accordance with one or more techniques of this disclosure.
[0022] FIG. 11 is a flowchart illustrating another example encoding
operation, in accordance with the techniques of this
disclosure.
[0023] FIG. 12 is a flowchart illustrating another example coding
operation, in accordance with the techniques of this
disclosure.
DETAILED DESCRIPTION
[0024] A video encoder may quantize coefficients during video
encoding. Quantizing the coefficients may reduce the bit depths of
the coefficients. In High Efficiency Video Coding (HEVC), the video
encoder may perform quantization using a scalar quantization with
deadzone technique or using a rate distortion optimized
quantization (RDOQ) technique. The RDOQ technique may achieve
significant performance gains over the scalar quantization with
deadzone technique. However, the RDOQ technique is significantly
more complex than the scalar quantization with deadzone
technique.
[0025] The techniques of this disclosure may reduce the complexity
of the RDOQ technique. For example, the RDOQ may involve
determining a rate distortion (RD) costs of a quantized level
(Level) of a coefficient, Level-1, and 0. Some techniques of this
disclosure may reduce the complexity of the RDOQ technique by only
determining a RD costs of a quantized level (Level), Level-1, and 0
for a coefficient if the calculated value of the coefficient is
less than the product of Level and a quantization step size value.
By determining these RD costs only when this condition is
satisfied, the complexity of performing the RDOQ technique may be
reduced. In another example, some techniques of this disclosure may
reduce the complexity of the RDOQ technique, in that a video
encoder implementing these techniques may look up quantization
offsets in a quantization offset lookup table and use these
quantization offsets to quantize coefficients. By looking up the
quantization offsets in the quantization offset lookup table, it
may be unnecessary to determine the RD costs of Level, Level-1, and
0 for each coefficient. Eliminating the determination of the RD
costs of Level, Level-1, and 0 may reduce the complexity of the
RDOQ technique. In another example, the complexity of the RDOQ
technique may be reduced by generating a multiplication table that
the video encoder may use when calculating various RD costs in the
RDOQ technique.
[0026] The attached drawings illustrate examples. Elements
indicated by reference numbers in the attached drawings correspond
to elements indicated by like reference numbers in the following
description. In this disclosure, elements having names that start
with ordinal words (e.g., "first," "second," "third," and so on) do
not necessarily imply that the elements have a particular order.
Rather, such ordinal words may merely be used to refer to different
elements of a same or similar type.
[0027] FIG. 1 is a block diagram illustrating an example video
coding system 10 that may utilize the techniques of this
disclosure. As described herein, the term "video coder" refers
generically to both video encoders and video decoders. In this
disclosure, the terms "video coding" or "coding" may refer
generically to video encoding or video decoding.
[0028] As shown in FIG. 1, video coding system 10 includes a source
device 12 and a destination device 14. Source device 12 generates
encoded video data. Accordingly, source device 12 may be referred
to as a video encoding device or a video encoding apparatus.
Destination device 14 may decode the encoded video data generated
by source device 12. Accordingly, destination device 14 may be
referred to as a video decoding device or a video decoding
apparatus. Source device 12 and destination device 14 may be
examples of video coding devices or video coding apparatuses.
Source device 12 and destination device 14 may comprise a wide
range of devices, including desktop computers, mobile computing
devices, notebook (e.g., laptop) computers, tablet computers,
set-top boxes, telephone handsets such as so-called "smart" phones,
televisions, cameras, display devices, digital media players, video
gaming consoles, in-car computers, or the like.
[0029] Destination device 14 may receive encoded video data from
source device 12 via a channel 16. Channel 16 may comprise one or
more media and/or devices capable of moving the encoded video data
from source device 12 to destination device 14. In one example,
channel 16 may comprise one or more communication media that enable
source device 12 to transmit encoded video data directly to
destination device 14 in real-time. In this example, source device
12 may modulate the encoded video data according to a communication
standard, such as a wireless communication protocol, and may
transmit the modulated video data to destination device 14. The one
or more communication media may include wireless and/or wired
communication media, such as a radio frequency (RF) spectrum or one
or more physical transmission lines. The one or more communication
media may form part of a packet-based network, such as a local area
network, a wide-area network, or a global network (e.g., the
Internet). The one or more communication media may include routers,
switches, base stations, or other equipment that facilitate
communication from source device 12 to destination device 14.
[0030] In another example, channel 16 may include a storage medium
that stores encoded video data generated by source device 12. In
this example, destination device 14 may access the storage medium
via disk access or card access. The storage medium may include a
variety of locally-accessed data storage media such as Blu-ray
discs, DVDs, CD-ROMs, flash memory, or other suitable digital
storage media for storing encoded video data.
[0031] In a further example, channel 16 may include a file server
or another intermediate storage device that stores encoded video
data generated by source device 12. In this example, destination
device 14 may access encoded video data stored at the file server
or other intermediate storage device via streaming or download. The
file server may be a type of server capable of storing encoded
video data and transmitting the encoded video data to destination
device 14. Example file servers include web servers (e.g., for a
website), file transfer protocol (FTP) servers, network attached
storage (NAS) devices, and local disk drives.
[0032] Destination device 14 may access the encoded video data
through a standard data connection, such as an Internet connection.
Example types of data connections include wireless channels (e.g.,
Wi-Fi connections), wired connections (e.g., DSL, cable modem,
etc.), or combinations of both that are suitable for accessing
encoded video data stored on a file server. The transmission of
encoded video data from the file server may be a streaming
transmission, a download transmission, or a combination of
both.
[0033] The techniques of this disclosure are not limited to
wireless applications or settings. The techniques may be applied to
video coding in support of a variety of multimedia applications,
such as over-the-air television broadcasts, cable television
transmissions, satellite television transmissions, streaming video
transmissions, e.g., via the Internet, encoding of video data for
storage on a data storage medium, decoding of video data stored on
a data storage medium, or other applications. In some examples,
video coding system 10 may be configured to support one-way or
two-way video transmission to support applications such as video
streaming, video playback, video broadcasting, and/or video
telephony.
[0034] In the example of FIG. 1, source device 12 includes a video
source 18, a video encoder 20, and an output interface 22. In some
examples, output interface 22 may include a modulator/demodulator
(modem) and/or a transmitter. Video source 18 may include a video
capture device, e.g., a video camera, a video archive containing
previously-captured video data, a video feed interface to receive
video data from a video content provider, and/or a computer
graphics system for generating video data, or a combination of such
sources of video data.
[0035] Video encoder 20 may encode video data from video source 18.
In some examples, source device 12 directly transmits the encoded
video data to destination device 14 via output interface 22. The
encoded video data may also be stored onto a storage medium or a
file server for later access by destination device 14 for decoding
and/or playback.
[0036] In the example of FIG. 1, destination device 14 includes an
input interface 28, a video decoder 30, and a display device 32. In
some examples, input interface 28 includes a receiver and/or a
modem. Input interface 28 may receive encoded video data over
channel 16. Display device 32 may be integrated with or may be
external to destination device 14. In general, display device 32
displays decoded video data. Display device 32 may comprise a
variety of display devices, such as a liquid crystal display (LCD),
a plasma display, an organic light emitting diode (OLED) display,
or another type of display device.
[0037] Video encoder 20 and video decoder 30 may operate according
to a video compression standard, such as the High Efficiency Video
Coding (HEVC) standard presently under development, and may conform
to a HEVC Test Model (HM). A draft of the upcoming HEVC standard,
referred to as "HEVC Working Draft 6" or "WD6," is described in
document JCTVC-H1003, Bross et al., "High efficiency video coding
(HEVC) text specification draft 6," Joint Collaborative Team on
Video Coding (JCT-VC) of ITU-T SG16 WP3 and ISO/IEC JTC1/SC29/WG11,
8th Meeting: San Jose, Calif., USA, February, 2012, which, as of
Feb. 21, 2013, is downloadable from
http://phenix.int-evry.fr/jct/doc_end_user/documents/8_San%20Jose/wg11/JC-
TVC-H1003-v22.zip, the entire content of which is incorporated
herein by reference. Another draft of the upcoming HEVC standard,
referred to as "HEVC Working Draft 9," is described in Bross et
al., "High Efficiency Video Coding (HEVC) text specification draft
9," Joint Collaborative Team on Video Coding (JCT-VC) of ITU-T SG16
WP3 and ISO/IEC JTC1/SC29/WG11, 11th Meeting: Shanghai, China,
October, 2012, which, as of Feb. 21, 2013, is downloadable from:
http://phenix.int-evry.fr/jct/doc_end_user/documents/11_Shanghai/wg11/JCT-
VC-K1003-v8.zip, the entire content of which is incorporated herein
by reference.
[0038] Alternatively, video encoder 20 and video decoder 30 may
operate according to other proprietary or industry standards,
including ITU-T H.261, ISO/IEC MPEG-1 Visual, ITU-T H.262 or
ISO/IEC MPEG-2 Visual, ITU-T H.263, ISO/IEC MPEG-4 Visual and ITU-T
H.264 (also known as ISO/IEC MPEG-4 AVC), including its Scalable
Video Coding (SVC) and Multiview Video Coding (MVC) extensions. The
techniques of this disclosure, however, are not limited to any
particular coding standard or technique.
[0039] Again, FIG. 1 is merely an example and the techniques of
this disclosure may apply to video coding settings (e.g., video
encoding or video decoding) that do not necessarily include any
data communication between encoding and decoding devices. In other
examples, data is retrieved from a local memory, streamed over a
network, or the like. An encoding device may encode and store data
to memory, and/or a decoding device may retrieve and decode data
from memory. In many examples, the encoding and decoding is
performed by devices that do not communicate with one another, but
simply encode data to memory and/or retrieve and decode data from
memory.
[0040] Video encoder 20 and video decoder 30 each may be
implemented as any of a variety of suitable circuitry, such as one
or more microprocessors, digital signal processors (DSPs),
application-specific integrated circuits (ASICs),
field-programmable gate arrays (FPGAs), discrete logic, hardware,
or any combinations thereof. If the techniques are implemented
partially in software, a device may store instructions for the
software in a suitable, non-transitory computer-readable storage
medium and may execute the instructions in hardware using one or
more processors to perform the techniques of this disclosure. Any
of the foregoing (including hardware, software, a combination of
hardware and software, etc.) may be considered to be one or more
processors. Each of video encoder 20 and video decoder 30 may be
included in one or more encoders or decoders, either of which may
be integrated as part of a combined encoder/decoder (CODEC) in a
respective device.
[0041] This disclosure may generally refer to video encoder 20
"signaling" certain information to another device, such as video
decoder 30. The term "signaling" may generally refer to the
communication of syntax elements and/or other data that represent
encoded video data. Such communication may occur in real- or
near-real-time. Alternately, such communication may occur over a
span of time, such as might occur when storing syntax elements to a
computer-readable storage medium in an encoded bitstream at the
time of encoding, which then may be retrieved by a decoding device
at any time after being stored to this medium.
[0042] As mentioned briefly above, video encoder 20 encodes video
data. The video data may comprise one or more pictures. Each of the
pictures may be a still image. In some instances, a picture may be
referred to as a video "frame." Video encoder 20 may generate a
bitstream that includes a sequence of bits that form a coded
representation of the video data. The coded representation of the
video data may include coded pictures and associated data. A coded
picture is a coded representation of a picture. The associated data
may include sequence parameter sets (SPSs), picture parameter sets
(PPSs), and other syntax structures. A SPS may contain parameters
applicable to zero or more sequences of pictures. A PPS may contain
parameters applicable to zero or more pictures. A syntax structure
may be a set of zero or more syntax elements present together in a
bitstream in a specified order.
[0043] To generate an encoded representation of a picture, video
encoder 20 may partition the picture into a grid of coding tree
blocks (CTBs). Thus, each CTB may be associated with a region of
the picture. In some instances, a CTB may be referred to as a "tree
block", a "largest coding unit" (LCU) or a "coding tree unit." The
CTBs of HEVC may be broadly analogous to the macroblocks of
previous standards, such as H.264/AVC. However, a CTB is not
necessarily limited to a particular size and may include one or
more coding units (CUs).
[0044] Each of the CTBs may be associated with a block of luminance
(luma) samples and two blocks of chrominance (chroma) samples that
correspond to pixels in the region associated with the CTB. For
ease of explanation, this disclosure may refer to a two-dimensional
array of samples as a sample block. Video encoder 20 may use
quad-tree partitioning to partition the region associated with a
CTB into regions associated with CUs, hence the name "coding tree
blocks."
[0045] The CTBs of a picture may be grouped into one or more
slices. In some examples, each of the slices includes an integer
number of CTBs. As part of encoding a picture, video encoder 20 may
generate encoded representations of each slice of the picture
(i.e., coded slices). To generate a coded slice, video encoder 20
may encode each CTB of the slice to generate encoded
representations of each of the CTBs of the slice (i.e., coded
CTBs).
[0046] To generate a coded CTB, video encoder 20 may recursively
perform quad-tree partitioning on the region associated with a CTB
to divide the region into progressively-smaller regions. The
smaller regions may be associated with CUs. Each of the smaller
regions may be associated with a block of luma samples and two
blocks of chroma samples. Thus, a CU may be associated with a block
of luma samples and two blocks of chroma samples. A partitioned CU
may be a CU whose region is partitioned into regions associated
with other CUs. A non-partitioned CU may be a CU whose region is
not partitioned into regions associated with other CUs.
[0047] Video encoder 20 may generate one or more prediction units
(PUs) for each non-partitioned CU. Each of the PUs of a CU may be
associated with a different region within the region of the CU.
Video encoder 20 may generate predictive sample blocks for each PU
of the CU.
[0048] Video encoder 20 may use intra prediction or inter
prediction to generate the predictive sample blocks for a PU. If
video encoder 20 uses intra prediction to generate the predictive
sample blocks of a PU, video encoder 20 may generate the predictive
sample blocks of the PU based on decoded samples of the picture
associated with the PU. If video encoder 20 uses inter prediction
to generate the predictive sample blocks of the PU, video encoder
20 may generate the predictive sample blocks of the PU based on
decoded samples of one or more pictures other than the picture
associated with the PU.
[0049] Video encoder 20 may generate residual sample blocks for a
CU based on predictive sample blocks of the PUs of the CU. The
residual sample blocks for the CU may indicate differences between
samples in the predictive sample blocks for the PUs of the CU and
corresponding samples in the original sample blocks of the CU.
[0050] Furthermore, as part of encoding a non-partitioned CU, video
encoder 20 may perform recursive quad-tree partitioning on the
residual sample blocks of the CU to partition the residual sample
blocks of the CU into one or more smaller residual sample blocks
associated with transform units (TUs) of the CU. Because the sample
blocks of the CU may include a block of luma samples and two blocks
of chroma samples, each of the TUs may be associated with a
residual sample block of luma samples and two residual sample
blocks of chroma samples.
[0051] Video coder 20 may apply one or more transforms to the
residual sample blocks associated with the TUs to generate
coefficient blocks (i.e., blocks of coefficients). Video encoder 20
may perform a quantization process on each of the coefficient
blocks. Quantization generally refers to a process in which
coefficients are quantized to possibly reduce the amount of data
used to represent the coefficients, providing further
compression.
[0052] Video encoder 20 may apply entropy encoding operations, such
as Context-Adaptive Binary Arithmetic Coding (CABAC) operations, to
at least some of these syntax elements. In addition to entropy
encoding the syntax elements of a coefficient block, video encoder
20 may apply inverse quantization and inverse transforms to the
transform block to reconstruct a residual sample block from the
transform block. Video encoder 20 may add the reconstructed
residual sample block to corresponding samples from one or more
predictive sample blocks to produce a reconstructed sample block.
By reconstructing sample blocks for each TU of a CU in this way,
video encoder 20 may reconstruct the sample blocks of the CU.
[0053] After video encoder 20 reconstructs the sample blocks of the
CU, video encoder 20 may perform a deblocking operation to reduce
blocking artifacts associated with the CU. After performing these
operations, video encoder 20 may store the reconstructed sample
blocks of the CU in a decoded picture buffer for use in generating
predictive sample blocks for PUs of other CUs.
[0054] Video decoder 30 may receive a bitstream. The bitstream may
include a coded representation of video data encoded by video
encoder 20. Video decoder 30 may parse the bitstream to extract
syntax elements from the bitstream. As part of extracting at least
some syntax elements from the bitstream, video decoder 30 may
entropy decode data in the bitstream.
[0055] Video decoder 30 may reconstruct the pictures of the video
data based on the syntax elements extracted from the bitstream. The
process to reconstruct the video data based on the syntax elements
may be generally reciprocal to the process performed by video
encoder 20 to generate the syntax elements. For instance, video
decoder 30 may generate, based on syntax elements associated with a
CU, predictive sample blocks for PUs of the CU. In addition, video
decoder 30 may inverse quantize coefficient blocks associated with
TUs of the CU. Video decoder 30 may perform inverse transforms on
the coefficient blocks to reconstruct residual sample blocks
associated with the TUs of the CU. Video decoder 30 may reconstruct
the sample blocks of a CU based on the predictive sample blocks and
the residual sample blocks.
[0056] As discussed above, video encoder 20 may generate residual
sample blocks. Each sample in a residual sample block may indicate
a difference between corresponding samples in an original version
of a sample block and a predictive version of the same sample
block. After generating a residual sample block, video encoder 20
may apply one or more transforms to the residual sample block. By
applying the one or more transforms to the residual sample block,
video encoder 20 may convert the residual sample block from a
sample domain to a frequency domain. The result is a block of
transform coefficients (i.e., a "transform coefficient block" or a
"coefficient block"). In some examples, the purpose of
transformation is to decorrelate the signal into different
frequency components. Furthermore, in some examples, video encoder
20 does not apply a transform to a residual sample block. In such
examples, video encoder 20 may treat the residual sample block in
the same way as a coefficient block. For ease of explanation,
discussion in this disclosure of coefficient blocks may apply to
residual sample blocks to which video encoder 20 does not apply the
one or more transforms.
[0057] Furthermore, as discussed above, after generating a
coefficient block, video encoder 20 may quantize the coefficients
of the coefficient block. Quantizing the coefficients may reduce
the bit depths of the coefficients. Quantization is the process
that may introduce distortion in encoding. After that the quantized
coefficients are encoded using variable length coding, such as
CABAC. In HEVC, there are two techniques for quantization. The
first quantization technique is scalar quantization with deadzone.
The second quantization technique is rate distortion optimized
quantization (RDOQ).
[0058] In the scalar quantization with deadzone technique, video
encoder 20 maps a coefficient to a discrete quantized level using
equation (1), below:
z = w + f .DELTA. sign ( W ) ( 1 ) ##EQU00001##
[0059] In equation (1), [] denotes rounding to the nearest integer
towards minus infinity and sign( ) is a function that returns the
sign of the input signal, W. .DELTA. is a quantization step and f
is the quantization offset. In equation (1), the quantization
offset f controls the size of the deadzone. The deadzone is the
interval that maps to 0. The quantization offset f also shifts the
reconstruction level away from the center of the interval that maps
to 0. z indicates the quantized level value.
[0060] During decoding, video decoder 30 may inverse quantize the
transform coefficient using equation (2), below:
W'=.DELTA.Z (2)
In equation (2), W' denotes a de-quantized coefficient, .DELTA. is
a quantization step, and Z is the received quantized level.
[0061] The RDOQ technique attempts to optimize the trade-offs
between rate (i.e., the bit rate of encoded video data) and
distortion. In general, video encoder 20 may perform the RDOQ
technique by performing the following steps. First, video encoder
20 may initialize a context: C1=1, C2=0. Second, video encoder 20
may scan through the coefficients of the coefficient block in a
reverse diagonal scan order. As video encoder 20 scans each
coefficient, video encoder 20 may use a quantization offset of 1/2
to quantize the coefficient. Third, video encoder 20 may determine
optimal levels for coefficients in the coefficient block. Fourth,
video encoder 20 may determine the optimal last significant
coefficient.
[0062] To determine the optimal levels for the coefficients in the
third step of the previous paragraph, video encoder 20 may perform
the following steps. First, video encoder 20 may determine the last
significant coefficient. When scanning the coefficients in a
reverse diagonal scan order, the last significant coefficient is
the first encountered coefficient that is quantized to a non-zero
level. Second, for each remaining non-zero coefficient, video
encoder 20 may choose an optimal level based on rate-distortion
(RD) costs of the quantized levels for the coefficient. The
remaining coefficients are coefficients that occur after the last
significant coefficient according to the reverse diagonal scan
order.
[0063] To choose the optimal quantized level for a coefficient
(i.e., the current coefficient), video encoder 20 may determine if
the quantized level for the current coefficient is less than 3. If
the quantized level for the current coefficient is less than 3,
video encoder 20 may calculate the RD cost that would result from
replacing the current coefficient with the value Level, where Level
denotes the quantized level for the current coefficient calculated
above. Furthermore, if the quantized level for the current
coefficient is less than 3, video encoder 20 may calculate the RD
cost that would result from replacing the current coefficient with
Level-1. In addition, if the quantized level for the current
coefficient is less than 3, video encoder 20 may calculate the RD
cost that would result from replacing the current coefficient with
the value 0.
[0064] If the quantized level for the current coefficient is
greater than or equal to 3, video encoder 20 may calculate the RD
cost that would result from replacing the current coefficient with
the value Level. Furthermore, if the quantized level for the
current coefficient is greater than or equal to 3, video encoder 20
may calculate the RD cost that would result from replacing the
current coefficient with the value Level-1.
[0065] Furthermore, when video encoder 20 chooses the optimal
quantized level for the current coefficient, video encoder 20 may
update the RD cost of quantizing all coefficients of the
coefficient block to 0 (resulting in an un-coded block). Video
encoder 20 also updates the RD cost of encoding each coefficient of
the coefficient block. In addition, video encoder 20 updates the
cost of encoding the significance map of the coefficient block. The
significance map of the coefficient block indicates which
coefficients of the coefficient block are non-zero.
[0066] In addition, video encoder 20 updates the context based on
the optimized quantized (L.sub.opt) of the current coefficient.
Video encoder 20 may determine L.sub.opt as the quantized level for
the current coefficient (i.e., Level, Level-1, or 0) that has a
lowest RD cost. Video encoder 20 may use the context to perform the
level decision of the next coefficient. For instance, video encoder
20 may determine, based at least in part on the context, various RD
costs associated with the next coefficient.
[0067] As indicated above, video encoder 20 may determine the
optimal last significant coefficient as part of performing the RDOQ
technique. To determine the optimal last significant coefficient,
video encoder 20 may perform the following steps. First, video
encoder 20 may initialize a value (d64BestCost) to be equal to the
RD cost if the current block is un-coded. Second, video encoder 20
may scan through the coefficients of the coefficient block
according to a reverse diagonal scan until video encoder 20
encounters coefficient whose quantized level is greater than 1.
[0068] When video encoder 20 scans a coefficient (i.e., the current
coefficient), video encoder 20 performs the following steps. First,
video encoder 20 may calculate a total block cost (totalCost) that
would result if the current coefficient is considered to be the
last significant coefficient. Second, if totalCost is less than
d64BestCost, video encoder 20 may set a best last position
indicator (iBestLastIdxP1) such that iBestLastIdxP1indicates the
position within the coefficient block of the current coefficient
and may set d64BestCost equal to totalCost.
[0069] In the RDOQ technique described above, video encoder 20
calculates several RD costs. Video encoder 20 may calculate a RD
cost as D+.lamda.*R. D denotes the mean square error between the
original coefficients of the coefficient block and the
reconstructed coefficients of the coefficient block. R denotes the
entropy of the corresponding context model. Video encoder 20 may
estimate R before performing RDOQ for a coefficient block or before
performing RDOQ for any coefficient block of a TU, the basic unit
of transform and quantization in HEVC. That is, the rate R of the
symbol may be estimated every time before RDOQ (or for every TU) as
the entropy of the corresponding context model.
[0070] In order to estimate the rate R (i.e., the rate to encode
the quantized level and related information such as significance
map and position of the last significant coefficient etc.) for
every TU, video encoder 20 may calculate the entropies of the CABAC
context models. In theory, CABAC can approximate the limit of
compression. That is, the average codeword length generated by
CABAC may, in theory, approximate the entropy of the source. When
video encoder 20 uses CABAC to encode a data symbol, the data
symbol is first binarized. The binarized data symbol may be
referred to as a "bin."
[0071] Video encoder 20 may then select a context model. The
context model may define the statistics of recently coded data
symbols. For example, in case of encoding the quantized level of 1,
the context model may be defined as the number of previous
coefficients that have been quantized to 1. In this way, a context
model may define a conditional probability for the current symbol.
That is, the context model may indicate the probability of the
current symbol, given the previously-coded symbols. Furthermore,
the context model may define the probability of a bin being "1" or
"0." When video encoder 20 performs CABAC, video encoder 20 may
encode the binarized data symbol (i.e., the bin) according to the
selected context model (i.e., "probability model").
[0072] After encoding the binarized data symbol, video encoder 20
updates the selected context model based at least in part on the
actual coded value. As long as CABAC encoding is ongoing, video
encoder 20 may continue updating the context model. Before video
encoder 20 performs RDOQ, video encoder 20 may load the current
probability (i.e., context) model of every context and video
encoder 20 may calculate the corresponding entropy. Video encoder
20 may use the entropy to estimate the number of bits to encode the
symbol for the corresponding context. In other words, video encoder
20 may use the entropy of a context to estimate how many bits would
be in an encoded version of the symbol if the context were used to
encode the symbol. Estimating the number of bits to encode the
symbol may be referred herein to as "bit estimation."
[0073] Thus, when video encoder 20 performs the RDOQ technique
described above, video encoder 20 may perform the following steps:
[0074] 1. Initialize context: C1=1, C2=0. [0075] 2. Quantize the
coefficients with an offset of 1/2 in reverse diagonal scan order;
[0076] 3. Decision of the optimal level. [0077] I. Fix last
significant coefficient: the first coefficient that is quantized to
non-zero in reverse diagonal scan. [0078] II. For the remaining
non-zero coefficients in the reverse diagonal scan, choose optimal
level based on RD cost. [0079] a. Choose the best level L.sub.opt.
[0080] i. If Level<3, calculate the RD cost
(D+.lamda.R.sub.Level) of Level, Level-1 and 0. [0081] ii. If
Level>=3, calculate the RD cost (D+.lamda.R.sub.Level) of Level,
Level-1. [0082] b. Update encoding cost. 'i. Cost of quantized to 0
(for uncoded block). [0083] ii. Cost of encoding each coefficient.
[0084] iii. Cost of encoding significance map
(.lamda.R.sub.sig.sub.--.sub.map). [0085] c. Update context based
on the decided level L.sub.opt. The context may be used by the
level decision of the next coefficient. [0086] III. Process the
next coefficient, and go to step II. [0087] 4. Decision of the
optimal last significant coefficient. [0088] 1) Initialize
d64BestCost to be the RD cost if the current block is uncoded.
[0089] 2) In reverse diagonal scan, calculate the total block cost
totalCost if the current coefficient is the last significant
coefficient. That is, .lamda.R.sub.positionX+.lamda.R.sub.positionY
need to be added and corresponding costs with significance map
should be adjusted. [0090] 3) If totalCost<d64BestCost, new best
last position (iBestLastIdxP1) is current position,
d64BestCost=totalCost. [0091] 4) Iterate until
LevelofCurCoeff>1.
[0092] Compared with the scalar quantization with deadzone
technique, the RDOQ technical may achieve significant performance
gains. However, these gains come at the expense of significantly
increased complexity for video encoder 20. Tables 1-4, below
illustrate HEVC performance RDOQ off (i.e., scalar quantization
with deadzone) versus RDOQ. As shown in Tables 1-4, compared with
scalar quantization with deadzone, RDOQ can achieve a performance
gain from 3.4% to 5.7%.
TABLE-US-00001 TABLE 1 All Intra HE Y U V Class A 5.6% -0.8% -2.4%
Class B 6.9% 0.5% 0.5% Class C 5.3% -0.4% 0.5% Class D 5.1% -1.8%
-0.8% Class E 5.1% 0.3% 0.9% Overall 5.7% -0.4% -0.3% 5.7% -0.2%
0.0% Enc Time[%] 85% Dec Time [%] 118%
TABLE-US-00002 TABLE 2 Low delay B HE Y U V Class A Class B 4.3%
5.8% 5.2% Class C 3.7% 5.7% 5.7% Class D 3.7% 6.3% 6.0% Class E
1.8% 3.4% 2.1% Overall 3.5% 5.5% 4.9% 3.5% 5.4% 4.9% Enc Time[%]
92% Dec Time [%] 81%
TABLE-US-00003 TABLE 3 Random Access HE Y U V Class A 5.3% 4.7%
5.1% Class B 5.6% 6.9% 6.3% Class C 5.3% 7.7% 7.7% Class D 5.1%
5.7% 6.0% Class E Overall 5.3% 6.3% 6.3% 5.6% 6.4% 6.4% Enc Time[%]
91% Dec Time [%] 104%
TABLE-US-00004 TABLE 4 Low Delay P HE Y U V Class A Class B 3.9%
5.3% 4.8% Class C 3.7% 6.0% 6.0% Class D 3.9% 6.4% 6.2% Class E
1.5% 3.7% 1.5% Overall 3.4% 5.5% 4.8% 3.2% 5.3% 4.8% Enc Time[%]
88% Dec Time [%] 83%
[0093] The techniques of this disclosure may simplify the step in
RDOQ of choosing an optimal quantized level for a coefficient (step
(3) in the description above). From the description of RDOQ above,
the third step (i.e., the decision of the optimal level), RD cost
of multiple levels may be calculated in order to choose the optimal
level. That is, video encoder 20 may choose the optimal level from
Level, Level-1 or 0 based on the RD cost in the third step of
RDOQ.
[0094] In accordance with the techniques of this disclosure, video
encoder 20 may calculate, for each coefficient of a coefficient
block, the RD costs of using Level, Level-1, and 0 as the quantized
levels for a coefficient, only if the value of the coefficient is
less than the reconstructed value of the coefficient. The
reconstructed value of the coefficient is equal to the quantized
level for the coefficient (Level) multiplied by the step size used
in generating the quantized level for the coefficient. In other
words, if video encoder 20 determines that the value of the
coefficient is greater than or equal to the reconstructed value of
the coefficient, video encoder 20 does not calculate RD costs for
Level, Level-1, and 0, but may instead simply select Level as the
quantized level for the coefficient, which may simplify the
selection of an optimal quantized level for the coefficient. In
other words, video encoder 20 may determine, without determining
rate-distortion costs for quantizing the coefficient to be Level,
Level-1, or 0 and in response to determining that the coefficient
is not less than the product of the initial quantized level (i.e.,
Level) and the quantization step size value, that the actual
quantized level for the coefficient is the initial quantized level
for the coefficient.
[0095] Calculating the RD costs of using Level, Level-1, and 0 as
the quantized levels for the coefficient only if the value of the
coefficient is less than the reconstructed value of the coefficient
may reduce the number of multiplication operations that the video
encoder performs. At the same time, the bitrate does not increase
excessively.
[0096] Thus, in accordance with the techniques of this disclosure,
video encoder 20 may perform the RDOQ technique of this disclosure
by performing the following steps. First, video encoder 20 may
initialize a context: C1=1, C2=0. Second, video encoder 20 may scan
through the coefficients of the coefficient block in a reverse
diagonal scan order. As video encoder 20 scans each coefficient,
video encoder 20 may quantize the coefficient with an offset of
1/2. Third, video encoder 20 may determine optimal levels for
coefficients in the coefficient block. Fourth, video encoder 20 may
determine the optimal last significant coefficient. Video encoder
20 may perform the first, second, and fourth steps of the RDOQ
technique as described above.
[0097] However, to determine the optimal levels for the
coefficients in the third step, video encoder 20 may perform the
following steps. First, video encoder 20 may determine the last
significant coefficient. When scanning the coefficients in a
reverse diagonal scan order (that is, generally proceeding from
highest-frequency coefficients to lowest-frequency coefficients),
the last significant coefficient is the first encountered
coefficient that is quantized to a non-zero level. Second, for each
remaining non-zero coefficient, video encoder 20 may choose an
optimal level based on RD costs of the quantized levels for the
coefficient. The remaining coefficients are coefficients that occur
after the last significant coefficient according to the reverse
diagonal scan order.
[0098] To choose the optimal quantized level for a coefficient
(i.e., the current coefficient), in accordance with the techniques
of this disclosure, video encoder 20 may perform the following
steps. First, video encoder 20 may determine whether the value of
the current coefficient is less than Level*StepSize, where Level is
the quantized level for the current coefficient determined in step
(2) and StepSize is the size of the quantization step .DELTA. used
to quantize the current coefficient. In other words, video encoder
20 may determine whether the value of the current coefficient is
less than the reconstructed value of the current coefficient. If
the value of the current coefficient is less than the reconstructed
value of the current coefficient, video encoder 20 may determine
whether the quantized level for the current coefficient is less
than 3.
[0099] In response to determining that the current coefficient is
less than 3, video encoder 20 may calculate the RD cost that would
result from replacing the current coefficient with the value Level,
where Level denotes the quantized level for the current
coefficient. Furthermore, in response to determining that the
current coefficient is less than 3, video encoder 20 may calculate
the RD cost that would result from replacing the current
coefficient with Level-1. In response to determining that the
current coefficient is less than 3, video encoder 20 may also
determine the RD cost that would result from replacing the current
coefficient with the value 0.
[0100] In response to determining that the quantized level for the
current coefficient is greater than or equal to 3, video encoder 20
may calculate the RD cost that would result from replacing the
current coefficient with the value Level. Furthermore, video
encoder 20 may calculate the RD cost that would result from
replacing the current coefficient with the value Level-1.
[0101] Video encoder 20 may select the optimal quantized level
L.sub.opt of the current coefficient as the version having the
lowest RD cost. Furthermore, video encoder 20 may update the RD
costs used in determining the optimal last significant coefficient.
That is, video encoder 20 may update the RD cost of quantizing all
coefficients of the coefficient block to 0 (resulting in an
un-coded block). In addition, video encoder 20 may update the cost
of encoding each coefficient of the coefficient block. Video
encoder 20 may also update the cost of encoding the significance
map of the coefficient block. Video encoder 20 may update the
context based on the determined optimal quantized level for the
current coefficient (L.sub.opt). Video encoder 20 may use the
context in the level decision of the next coefficient in the
coefficient block.
[0102] In this way, video encoder 20 may determine an initial
quantized level for a coefficient of a coefficient block. In
addition, video encoder 20 may determine whether the coefficient is
less than the product of the initial quantized level and a
quantization step size value. In response to determining that the
transform coefficient is less than the product of the initial
quantized level and the quantization step size value, video encoder
20 may determine rate-distortion costs of quantizing the
coefficient to be the initial quantized level and the initial
quantized level minus one. Furthermore, in response to determining
that the transform coefficient is less than the product of the
initial quantized level and the quantization step size value, video
encoder 20 may determine an actual quantized level for the
coefficient based on the calculated rate-distortion costs. In
addition, video encoder 20 may include the actual quantized level
in a quantized version of the coefficient block.
[0103] As described above, the RDOQ algorithm has two main passes.
The first pass is to choose the optimal quantized level from among
Level, Level-1, and 0, based on the RD cost of the individual
coefficient. For a fixed quantization step, video encoder 20 may
always choose a quantization offset so that the quantized level is
equal to the level determined by the RDOQ algorithm. The second
pass is to select the optimal last significant coefficient based on
the RD cost of the block. Additional techniques of this disclosure
may simplify the RDOQ algorithm by replacing the first step of the
RDOQ algorithm. That is, the additional techniques of this
disclosure may simplify the step in RDOQ of choosing an optimal
quantized level for a coefficient. In particular, if the
quantization offset is adjusted so that the quantized level is
equal to the optimal level determined in the first pass of RDOQ,
then the complexities involved in the level decision of the first
pass may be removed.
[0104] To determine quantization offsets that can approximate the
performance of RDOQ, the number of times Level, Level-1 or 0 are
selected in steps 3 and 4 (in last significant coefficient
decision, some levels are set to 0) of RDOQ can be collected for
each frequency coefficient. In this way, the rank of Level, Level-1
or 0 being selected may be obtained for each coefficient at each
different frequency. If the quantization offsets are limited to
1/2, 1/3, and 1/6, video encoder 20 may select a quantization
offset as shown in Table 5, below. Table 5, below, illustrates an
example decision of quantization offset based on the rank of the
number of times Level, Level-1, or 0 get selected in RDOQ. In some
examples, Table 5 is generated based on test video data. In other
examples, Table 5 is generated dynamically by video encoder 20.
TABLE-US-00005 TABLE 5 Rank #1 Quantization offset Level 1/2
Level-1 1/3 0 1/6
For example, if Level is selected most frequently for a coefficient
at position (I,J) of a coefficient block, the quantization offset
1/2 may be set for the position (I,J). Otherwise if Level-1 or 0 is
selected most frequently in RDOQ, then quantization offset may be
set to 1/3 or 1/6. In this way, a near optimal quantization offset
table can be generated.
[0105] Thus, in accordance with the additional techniques of this
disclosure, a lookup table is generated. In some examples, video
encoder 20 generates the lookup table. The lookup table may map
transform coefficient positions to quantization offsets. The lookup
table may be generated based on rates at which the conventional
RDOQ technique uses the quantization offsets for coefficients at
various positions within coefficient blocks. Example quantization
offsets may include 1/2, 1/3, and 1/6. That is, the lookup table
may map a coefficient position to one of three possible
quantization offset values: 1/2, 1/3, and 1/6. Video encoder 20 may
use respective positions of respective coefficients in a
coefficient block to look up, in the lookup table, respective
quantization offsets for the respective coefficients. Furthermore,
video encoder 20 may determine, based at least in part on the
respective quantization offsets for the one or more respective
coefficients, respective quantized levels for the respective
coefficients.
[0106] After video encoder 20 uses the lookup table to look up a
quantization offset for a coefficient (i.e., the current
coefficient), video encoder 20 may use equation (1), reproduced
again below, to determine the optimized quantized level (L.sub.opt)
for the current coefficient.
z = w + f .DELTA. sign ( W ) ( 1 ) ##EQU00002##
In other words, after determining whether to use value Level,
Level-1, or 0, video encoder 20 may select a value for f in
equation (1) above according to a table similar to Table 5. In this
manner, video encoder 20 may calculate a quantized level for the
current coefficient (i.e., the value of z calculated according to
function (1)). Video encoder 20 may use the quantized level for the
current coefficient in the quantized coefficient block.
[0107] In the RDOQ technique, video encoder 20 computes the RD
cost. Computing the RD cost may involve a number of double point
additions and multiplications. Double point addition and
multiplication are operations that involve the addition and
multiplication of numbers that involve the double-precision type of
data (e.g., a floating point number). Double point multiplication
may take more computing time than other types of computing
operations. Consequently, the major computations for the RDOQ
technique for each coefficient may occur in the third step (i.e.,
the decision of the optimal level for the coefficient). The double
point multiplications in the third step of the RDOQ technique
include: [0108] Two double point multiplications for reconstruction
error D. D=dError*dError*dTemp, where dError is the difference of
the original coefficient and the reconstructed coefficient and
dTemp is a scaling factor. [0109] One double point multiplication
for .lamda.R.sub.level, where R.sub.level is the number of bits
estimated to encode a quantized level. [0110] One double point
multiplication for .lamda.R.sub.sig.sub.--.sub.map, where
R.sub.sig.sub.--.sub.map is the number of bits estimated to encode
the significance map for the current coefficient. [0111] Two double
point multiplications for distortion induced if the current
coefficient is quantized to 0. Furthermore, the major computations
in the fourth step of the RDOQ technique (i.e., the decision of the
optimal last significant coefficient) include two double point
multiplications for .lamda.R.sub.positionX+.lamda.R.sub.positionY,
where R.sub.positionX and R.sub.positionY are the number of bits
estimated to encode the position of the last significant
coefficient. Although is a constant number, video encoder 20 may
have to calculate .lamda.R as R is estimated as the entropy of the
updating context model.
[0112] Temporally neighboring pictures are typically highly
correlated. That is, neighboring pictures are likely to be similar
to one another. Hence, in accordance with one or more techniques of
this disclosure, video encoder 20 may use the entropies of the
context models of the previous frame as the bit estimation for the
symbols in the current frame. In this way, it may be unnecessary
for video encoder 20 to load the entropies of the context models
each time video encoder 20 performs RDOQ. Similarly, it may be
unnecessary for video encoder 20 to load the entropies of the
context models for every TU. Rather, video encoder 20 may load the
entropies of the context model each time video encoder 20 starts
encoding a new frame. In other words, video encoder 20 may perform
bit estimation once per frame. In other words, video encoder 20 may
estimate number of bits for the symbols in the current frame once
per frame.
[0113] Because .lamda. is a constant number, video encoder 20 may
avoid the double point multiplications by constructing a
multiplication table at the beginning of encoding each new frame.
The multiplication table may be a lookup table. Video encoder 20
may use the context as an index to lookup a value of .lamda.R in
the multiplication table. During RD cost calculation in RDOQ, video
encoder 20 may look up the RD cost in the multiplication table. The
bit estimation in RD cost may not be as accurate because there may
be mismatches between the current context models and the context
models of the previous frame. However, due to the fact that
temporally neighboring video frames are highly correlated, the
performance loss brought by the mismatch may be relatively
small.
[0114] FIG. 2 is a block diagram illustrating an example video
encoder that may implement the techniques described in this
disclosure. FIG. 2 is provided for purposes of explanation and
should not be considered limiting of the techniques as broadly
exemplified and described in this disclosure. For purposes of
explanation, this disclosure describes video encoder 20 in the
context of HEVC coding. However, the techniques of this disclosure
may be applicable to other coding standards or methods.
[0115] In the example of FIG. 2, video encoder 20 includes a
prediction processing unit 100, a residual generation unit 102, a
transform processing unit 104, a quantization unit 106, an inverse
quantization unit 108, an inverse transform processing unit 110, a
reconstruction unit 112, a filter unit 113, a decoded picture
buffer 114, and an entropy encoding unit 116. Entropy encoding unit
116 includes a regular CABAC coding engine 118 and a bypass coding
engine 120. Prediction processing unit 100 includes an
inter-prediction processing unit 121 and an intra-prediction
processing unit 126. Inter-prediction processing unit 121 includes
a motion estimation unit 122 and a motion compensation unit 124. In
other examples, video encoder 20 may include more, fewer, or
different functional components.
[0116] Video encoder 20 receives video data. As part of encoding a
picture, video encoder 20 may encode each coding tree block (CTB)
in a slice. As part of encoding a CTB, prediction processing unit
100 may perform quad-tree partitioning on the region associated
with the CTB to divide the region into progressively-smaller
regions. The smaller regions may be associated with CUs. For
example, prediction processing unit 100 may partition the region of
a CTB into four equally-sized sub-regions, partition one or more of
the sub-regions into four equally-sized sub-sub-regions, and so
on.
[0117] Video encoder 20 may encode CUs of a CTB in a picture to
generate encoded representations of the CUs (i.e., coded CUs).
Video encoder 20 may encode the CUs of a CTB according to a z-scan
order. In other words, video encoder 20 may encode a top-left CU, a
top-right CU, a bottom-left CU, and then a bottom-right CU, in that
order. When video encoder 20 encodes a partitioned CU, video
encoder 20 may encode CUs associated with sub-blocks of the sample
blocks of the partitioned CU according to the z-scan order.
[0118] Furthermore, as part of encoding a CU, prediction processing
unit 100 may partition the sample blocks of the CU among one or
more PUs of the CU. Video encoder 20 and video decoder 30 may
support various PU sizes. Assuming that the size of a particular CU
is 2N.times.2N, video encoder 20 and video decoder 30 may support
PU sizes of 2N.times.2N or N.times.N for intra prediction, and
symmetric PU sizes of 2N.times.2N, 2N.times.N, N.times.2N,
N.times.N, or similar for inter prediction. Video encoder 20 and
video decoder 30 may also support asymmetric partitioning for PU
sizes of 2N.times.nU, 2N.times.nD, nL.times.2N, and nR.times.2N for
inter prediction.
[0119] Inter-prediction processing unit 121 may generate predictive
data for a PU by performing inter prediction on each PU of a CU.
The predictive data for the PU may include predictive sample blocks
that correspond to the PU and motion information for the PU. Slices
may be I slices, P slices, or B slices. Inter-prediction unit 121
may perform different operations for a PU of a CU depending on
whether the PU is in an I slice, a P slice, or a B slice. In an I
slice, all PUs are intra predicted. Hence, if the PU is in an I
slice, inter-prediction unit 121 does not perform inter prediction
on the PU.
[0120] If a PU is in a P slice, motion estimation unit 122 may
search the reference pictures in a list of reference pictures
(e.g., "list 0") for a reference block for the PU. Data for the
reference pictures may be stored in decoded picture buffer 114. The
reference block of the PU may be a sample block that most closely
corresponds to a sample block of the PU. Motion estimation unit 122
may generate a reference picture index that indicates the reference
picture in list 0 containing the reference block of the PU and a
motion vector that indicates a spatial displacement between a
sample block of the PU and the reference block. Motion estimation
unit 122 may output the reference picture index and the motion
vector as the motion information of the PU. Motion compensation
unit 124 may generate the predictive sample blocks of the PU based
at least in part on the reference block indicated by the motion
information of the PU.
[0121] If a PU is in a B slice, motion estimation unit 122 may
perform uni-directional inter prediction or bi-directional inter
prediction for the PU. To perform uni-directional inter prediction
for the PU, motion estimation unit 122 may search the reference
pictures of a first reference picture list ("list 0") or a second
reference picture list ("list 1") for a reference block for the PU.
Motion estimation unit 122 may output, as the motion information of
the PU, a reference picture index that indicates a position in list
0 or list 1 of the reference picture that contains the reference
block, a motion vector that indicates a spatial displacement
between a sample block of the PU and the reference block, and a
prediction direction indicator that indicates whether the reference
picture is in list 0 or list 1.
[0122] To perform bi-directional inter prediction for a PU, motion
estimation unit 122 may search the reference pictures in list 0 for
a reference block for the PU and may also search the reference
pictures in list 1 for another reference block for the PU. Motion
estimation unit 122 may generate reference picture indexes that
indicate positions in list 0 and list 1 of the reference pictures
that contain the reference blocks. In addition, motion estimation
unit 122 may generate motion vectors that indicate spatial
displacements between the reference blocks and the sample blocks of
the PU. The motion information of the PU may include the reference
picture indexes and the motion vectors of the PU. Motion
compensation unit 124 may generate the predictive sample blocks of
the PU based on the reference blocks indicated by the motion
information of the PU.
[0123] Intra-prediction processing unit 126 may generate predictive
data for a PU by performing intra prediction on the PU. The
predictive data for the PU may include predictive sample blocks for
the PU and various syntax elements. Intra-prediction processing
unit 126 may perform intra prediction on PUs in I slices, P slices,
and B slices.
[0124] To perform intra prediction on a PU, intra-prediction
processing unit 126 may use multiple intra prediction modes to
generate multiple sets of predictive data for the PU. To use an
intra prediction mode to generate a set of predictive data for the
PU, intra-prediction processing unit 126 may extend samples from
sample blocks of neighboring PUs across the sample blocks of the PU
in a direction associated with the intra prediction mode. The
neighboring PUs may be above, above and to the right, above and to
the left, or to the left of the PU, assuming a left-to-right,
top-to-bottom encoding order for PUs, CUs, and CTBs.
Intra-prediction processing unit 126 may use various numbers of
intra prediction modes, e.g., 33 directional intra prediction
modes. In some examples, the number of intra prediction modes may
depend on the size of the sample blocks of the PU.
[0125] Prediction processing unit 100 may select the predictive
data for PUs of a CU from among the predictive data generated by
inter-prediction processing unit 121 for the PUs or the predictive
data generated by intra-prediction processing unit 126 for the PUs.
In some examples, prediction processing unit 100 selects the
predictive data for the PUs of the CU based on rate/distortion
metrics of the sets of predictive data. The predictive sample
blocks of the selected predictive data may be referred to herein as
the selected predictive sample blocks.
[0126] Residual generation unit 102 may generate, based at least in
part on the sample blocks of a CU and the selected predictive
sample blocks of the PUs of the CU, residual sample blocks of a CU.
For instance, residual generation unit 102 may generate the
residual sample blocks of the CU such that each sample in the
residual sample blocks has a value equal to a difference between a
sample in a sample block of the CU and a corresponding sample in a
selected predictive sample block of a PU of the CU.
[0127] Prediction processing unit 100 may perform quad-tree
partitioning to partition the residual sample blocks of a CU into
sub-blocks. Each undivided residual sample block may be associated
with TUs of the CU. The sizes and positions of the residual sample
blocks associated with TUs of a CU may or may not be based on the
sizes and positions of sample blocks of the PUs of the CU.
[0128] Because the CU is associated with one block of luma samples
and two blocks of chroma samples, each of the TUs may be associated
with a block of luma samples and two blocks of chroma samples.
Transform processing unit 104 may generate coefficient blocks for
each TU of a CU by applying one or more transforms to the residual
sample blocks associated with the TU. Transform processing unit 104
may apply various transforms to a residual sample block associated
with a TU. For example, transform processing unit 104 may apply a
discrete cosine transform (DCT), a directional transform, or a
conceptually-similar transform to a residual sample block.
[0129] Quantization unit 106 may quantize the coefficients in a
coefficient block. The quantization process may reduce the bit
depth associated with some or all of the coefficients. For example,
an n-bit coefficient may be rounded down to an m-bit coefficient
during quantization, where n is greater than m. Quantization unit
106 may quantize a coefficient block associated with a TU of a CU
based at least in part on a quantization parameter (QP) value
associated with the CU. Video encoder 20 may adjust the degree of
quantization applied to the coefficient blocks associated with a CU
by adjusting the QP value associated with the CU. In accordance
with the techniques of this disclosure, quantization unit 106 may
be configured to select a quantized level value for a coefficient
using a look-up table and/or according to a modified RDOQ process,
as explained above with respect to FIG. 1.
[0130] Inverse quantization unit 108 and inverse transform
processing unit 110 may apply inverse quantization and inverse
transforms to a coefficient block, respectively, to reconstruct a
residual sample block from the coefficient block. Reconstruction
unit 112 may add samples of the reconstructed residual sample block
to corresponding samples from one or more predictive sample blocks
generated by prediction processing unit 100 to produce a
reconstructed sample block associated with a TU. By reconstructing
sample blocks for each TU of a CU in this way, video encoder 20 may
reconstruct the sample blocks of the CU.
[0131] Filter unit 113 may perform a deblocking operation to reduce
blocking artifacts in the sample blocks associated with a CU.
Decoded picture buffer 114 may store the reconstructed sample
blocks. Inter-prediction unit 121 may use a reference picture that
contains the reconstructed sample blocks to perform inter
prediction on PUs of other pictures. In addition, intra-prediction
processing unit 126 may use reconstructed sample blocks in decoded
picture buffer 114 to perform intra prediction on other PUs in the
same picture as the CU.
[0132] Entropy encoding unit 116 may receive data from other
functional components of video encoder 20. For example, entropy
encoding unit 116 may receive coefficient blocks from quantization
unit 106 and may receive syntax elements from prediction processing
unit 100. Entropy encoding unit 116 may perform one or more entropy
encoding operations on the data to generate entropy-encoded data.
For example, entropy encoding unit 116 may perform a
context-adaptive variable length coding (CAVLC) operation, a CABAC
operation, a variable-to-variable (V2V) length coding operation, a
syntax-based context-adaptive binary arithmetic coding (SBAC)
operation, a Probability Interval Partitioning Entropy (PIPE)
coding operation, or another type of entropy encoding operation on
the data. Video encoder 20 may output a bitstream that includes
entropy-encoded data generated by entropy encoding unit 116.
[0133] FIG. 3 is a block diagram illustrating an example video
decoder 30 that may implement the techniques of this disclosure.
FIG. 3 is provided for purposes of explanation and is not limiting
on the techniques as broadly exemplified and described in this
disclosure. For purposes of explanation, this disclosure describes
video decoder 30 in the context of HEVC coding. However, the
techniques of this disclosure may be applicable to other coding
standards or methods.
[0134] In the example of FIG. 3, video decoder 30 includes an
entropy decoding unit 150, a prediction processing unit 152, an
inverse quantization unit 154, an inverse transform processing unit
156, a reconstruction unit 158, a filter unit 159, and a decoded
picture buffer 160. Prediction processing unit 152 includes a
motion compensation unit 162 and an intra-prediction processing
unit 164. Entropy decoding unit 150 includes a regular CABAC coding
engine 166 and a bypass coding engine 168. In other examples, video
decoder 30 may include more, fewer, or different functional
components.
[0135] Video decoder 30 may receive a bitstream. Entropy decoding
unit 150 may parse the bitstream to extract syntax elements from
the bitstream. In addition, entropy decoding unit 150 may entropy
decode entropy-encoded syntax elements in the bitstream. Prediction
processing unit 152, inverse quantization unit 154, inverse
transform processing unit 156, reconstruction unit 158, and filter
unit 159 may generate decoded video data based on the syntax
elements extracted from the bitstream.
[0136] In addition, video decoder 30 may perform a reconstruction
operation on a non-partitioned CU. To perform the reconstruction
operation on a non-partitioned CU, video decoder 30 may perform a
reconstruction operation on each TU of the CU. By performing the
reconstruction operation for each TU of the CU, video decoder 30
may reconstruct residual sample blocks of the CU.
[0137] As part of performing a reconstruction operation on a TU of
a CU, inverse quantization unit 154 may inverse quantize, i.e.,
de-quantize, coefficient blocks associated with the TU. Inverse
quantization unit 154 may use a QP value associated with the CU of
the TU to determine a degree of quantization and, likewise, a
degree of inverse quantization for inverse quantization unit 154 to
apply. Inverse quantization unit 154 may inverse quantize a
coefficient according to a conventional inverse quantization
process, and need not utilize the RDOQ process or a look-up table
to determine inverse quantized values for coefficients. In this
manner, the techniques of this disclosure may be used to produce a
standard-compliant bitstream that video decoder 30 can decode
without modification.
[0138] After inverse quantization unit 154 inverse quantizes a
coefficient block, inverse transform processing unit 156 may apply
one or more inverse transforms to the coefficient block in order to
generate a residual sample block associated with the TU. For
example, inverse transform processing unit 156 may apply an inverse
DCT, an inverse integer transform, an inverse Karhunen-Loeve
transform (KLT), an inverse rotational transform, an inverse
directional transform, or another inverse transform to the
coefficient block.
[0139] If a PU is encoded using intra prediction, intra-prediction
processing unit 164 may perform intra prediction to generate a
predictive sample block for the PU. Intra-prediction processing
unit 164 may use an intra prediction mode to generate the
predictive sample blocks for the PU based on the sample blocks of
spatially-neighboring PUs. Intra-prediction processing unit 164 may
determine the intra prediction mode for the PU based on one or more
syntax elements parsed from the bitstream.
[0140] Motion compensation unit 162 may construct a first reference
picture list (list 0) and a second reference picture list (list 1)
based at least in part on syntax elements extracted from the
bitstream. Furthermore, if a PU is encoded using inter prediction,
entropy decoding unit 150 may extract motion information for the
PU. Motion compensation unit 162 may determine, based on the motion
information of the PU, one or more reference blocks for the PU.
Motion compensation unit 162 may generate, based at least in part
on the one or more reference blocks for the PU, predictive sample
blocks for the PU.
[0141] Reconstruction unit 158 may use the residual sample blocks
associated with TUs of a CU and the predictive sample blocks of the
PUs of the CU, i.e., either intra-prediction data or
inter-prediction data, as applicable, to reconstruct the sample
blocks of the CU. In particular, reconstruction unit 158 may add
samples of the residual sample blocks to corresponding samples of
the predictive sample blocks to reconstruct the sample blocks of
the CU.
[0142] Filter unit 159 may perform a deblocking operation to reduce
blocking artifacts associated with the sample blocks of the CUs of
a CTB. Video decoder 30 may store the sample blocks of the CU in
decoded picture buffer 160. Decoded picture buffer 160 may provide
reference pictures for subsequent motion compensation, intra
prediction, and presentation on a display device, such as display
device 32 of FIG. 1. For instance, video decoder 30 may perform,
based at least in part on the sample blocks in decoded picture
buffer 160, intra prediction or inter prediction operations on PUs
of other CUs.
[0143] FIG. 4 is a conceptual diagram illustrating scalar
quantization with deadzone. In the example of FIG. 4, each circle
illustrates a reconstruction value W' and each vertical line
indicates a decision level. As can be seen from FIG. 4, the
quantization offset f controls the size of the deadzone (i.e., the
interval that maps to 0). The quantization offset f also shifts the
reconstruction level away from the center of the interval.
[0144] The example of FIG. 4 shows the cases where f=.DELTA./2 and
f=.DELTA./3, where f represents the quantization offset in
accordance with formula (1) described above. It can be observed
from the example of FIG. 4 that when the quantization offset is
decreased from f=.DELTA./2 to f=.DELTA./3, the deadzone increases
and the reconstructed level shifts toward 0. To minimize the
reconstruction error, the reconstructed value may be equal to the
mean of the interval with respect to the distribution of the
transformed coefficients. The distribution of the transformed
coefficients may be modeled as a Laplacian distribution. That is,
the quantization offset f that gives the minimal reconstruction
error should be less than .DELTA./2. In addition, by reducing the
quantization offset f lower than .DELTA./2, coefficients originally
quantized to level n may be quantized to n-1. In HEVC, the
quantization offset f is empirically set to be .DELTA./3 for
intra-mode and .DELTA./6 for inter-mode.
[0145] FIG. 5 is a conceptual diagram illustrating quantization
offset and level decision. In other words, FIG. 5 shows a
relationship between a quantization offset and a resulting level.
In the example of FIG. 5, vertical bars 250 indicate the decision
level when the quantization offset is .DELTA./2. As the
quantization offset changes (shown in FIG. 5 as dashed vertical
bars), the value may be quantized as level n or level n-1 depending
on where the value falls. If the value falls in the range from
n.DELTA. to (n+1/2).DELTA., no matter what the quantization offset
is, the quantized level may still be n, as illustrated in FIG. 5 by
zone 252. However, if the value falls in the range from
(n-1/2).DELTA. to n.DELTA. (zone 254), then depending on the
quantization offset and where the value falls, the quantization
level can be n or n-1.
[0146] FIG. 6 is a flowchart illustrating an example operation 300
to quantize a coefficient, in accordance with one or more
techniques of this disclosure. In the example of FIG. 6,
quantization unit 106 may determine an initial quantized level for
a coefficient of a coefficient block (302). The coefficient may
occur after a first non-zero coefficient in a reverse scan order in
the coefficient block. Furthermore, quantization unit 106 may
determine whether the coefficient is less than the product of the
initial quantized level and a quantization step size value
(304).
[0147] In response to determining that the transform coefficient is
less than the product of the initial quantized level and the
quantization step size value ("YES" of 304), quantization unit 106
may determine whether the initial quantized level is less than
three (305). In response to determining that the initial quantized
level is less than three ("YES" of 305), rate-distortion costs of
quantizing the coefficient to be the initial quantized level for
the coefficient, the initial quantized level minus one, and zero
(306). Thus, quantization unit 106 may determine, in response to
determining that the initial quantized level is less than three,
rate-distortion costs of quantizing the coefficient to be the
initial quantized level for the coefficient, the initial quantized
level minus one, and zero. Furthermore, in response to determining
that the initial quantized level for the coefficient is not less
than three ("NO" of 305), quantization unit 106 may determine
rate-distortion costs of quantizing the coefficient to be the
initial quantized level and the initial quantized level minus one
(308). That is, quantization unit 106 may determine, in response to
determining that the initial quantized level is greater than or
equal to three, rate-distortion costs of quantizing the coefficient
to be the initial quantized level and the initial quantized level
minus one. More generally, quantization unit 106 may determine
whether the coefficient is less than the product of the initial
quantized level and a quantization step size value. In response to
determining that the transform coefficient is less than the product
of the initial quantized level and the quantization step size
value, quantization unit 106 may determine rate-distortion costs of
quantizing the coefficient to be the initial quantized level and
the initial quantized level minus one.
[0148] After performing action 306 or action 308, quantization unit
106 may determine an actual quantized level for the coefficient
based, at least in part, on the calculated rate-distortion costs
(310). After performing actions 306 or 308 and 310, or in response
to determining that the coefficient is not less than the product of
the initial quantized level and the quantization step size value
("NO" of 304), quantization unit 106 may include the actual
quantized level in a quantized version of the coefficient block
(312). If the coefficient is not less than the product of the
initial quantized level and the quantization step size value, the
actual quantized level may be the initial quantized level for the
coefficient. In addition, quantization unit 106 may entropy encode
(e.g., CABAC encode) the actual quantized level (314). Quantization
unit 106 may output a bitstream that includes the entropy-encoded
actual quantized level (316).
[0149] FIG. 7 is a flowchart illustrating an example operation 350
to quantize coefficients of a coefficient block, in accordance with
one or more techniques of this disclosure. Operation 350 of FIG. 7
may implement techniques similar to those of operation 300 of FIG.
6. In the example of FIG. 7, quantization unit 106 may initialize a
context (352) such as C1=1, C2=0. Quantization unit 106 may use the
context in determining RD costs. Furthermore, quantization unit 106
may scan through the coefficients of the coefficient block in a
reverse diagonal scan order and quantize the coefficients in the
coefficient block using a quantization offset of 1/2 (354). Next,
quantization unit 106 may determine a last significant coefficient
of the coefficient block (356). The last significant coefficient
may be the first non-zero transform coefficient in the reverse scan
order. When scanning the coefficients in a reverse diagonal scan
order (that is, generally proceeding from highest-frequency
coefficients to lowest-frequency coefficients), the last
significant coefficient is the first encountered coefficient that
is quantized to a non-zero level.
[0150] After determining the last significant coefficient of the
coefficient block, quantization unit 106 may perform the portion of
operation 350 illustrated in FIG. 8. FIG. 8 is described in detail
later in this disclosure. When quantization unit 106 performs the
portion of operation 350 illustrated in FIG. 8, quantization unit
106 may determine, for each respective remaining non-zero
coefficient, an optimal level for the respective remaining non-zero
coefficient. As described below with regard to FIG. 8, quantization
unit 106 may determine the optimal level for the respective
remaining non-zero coefficient based at least in part on RD costs
of quantized levels for the respective remaining non-zero
coefficient. The remaining coefficients are coefficients that occur
after the last significant coefficient according to the reverse
diagonal scan order.
[0151] After performing the portion of operation 350 illustrated in
FIG. 8, quantization unit 106 may initialize a best code value to
be equal to an RD cost if the current block is un-coded (358). That
is, quantization unit 106 may determine the RD cost that would
result from quantizing each coefficient of the coefficient block to
0. Quantization unit 106 may set the best code value to be equal to
this RD cost.
[0152] In addition, quantization unit 106 may initialize a current
coefficient indicator (360). In some examples, quantization unit
106 may initialize the current coefficient indicator to indicate
the last coefficient (e.g., the bottom right coefficient) of the
coefficient block. The current coefficient indicator indicates a
coefficient in the coefficient block. This disclosure may refer to
the coefficient indicated by the current coefficient indicator as
the current coefficient.
[0153] Next, quantization unit 106 may determine whether the
current coefficient has a quantized level greater than 1 (362). In
response to determining that the current coefficient has a
quantized level greater than 1 ("YES" of 362), quantization unit
106 may end operation 350. On the other hand, in response to
determining that the current coefficient does not have a quantized
level greater than 1 ("NO" of 362), quantization unit 106 may
calculate a total block RD cost that would result if the current
coefficient is the last significant coefficient (364). Here, the
total block RD cost may be calculated as follows
.SIGMA..sup.N.sub.i=1
Distortion(C.sub.i)+.lamda.(.SIGMA..sup.N.sub.i=1
(R.sub.i+R.sub.pos.sub.x+R.sub.pos.sub.y+R.sub.significance.sub.--.sub.ma-
p)) where N is the total number of the coefficients in the current
TU, C.sub.i is the coefficient i in TU, .lamda. is a fixed value,
R.sub.i the number of the bits to encode the quatized level of the
coefficient, C.sub.i, R.sub.pos.sub.x, R.sub.pos.sub.y are the
number of bits to encode the position x, and y for the last
significant coefficients, and R.sub.significance.sub.--.sub.map is
the number of bits to encode the significance map.
[0154] Furthermore, quantization unit 106 may determine whether the
calculated total block RD cost is less than the best cost value
(366). In response to determining that the calculated total block
RD cost is less than the best cost value ("YES" of 366),
quantization unit 106 may set a best last position indicator to the
current position and may set the best cost value to the calculated
total block RD cost (368). The current position is the position
within the coefficient block of the current coefficient.
[0155] After setting the best last position indicator and the best
cost value or after determining that the calculated total block RD
cost is not less than the best cost value ("NO" of 366),
quantization unit 106 may update the current coefficient indicator
such that the current coefficient indicator indicates a next
coefficient in the coefficient block, according to the reverse
diagonal scan order (370). Quantization unit 106 may then determine
again whether the current coefficient has a quantized level greater
than 1 (362). In this way, quantization unit 106 may scan through
the coefficients of the coefficient block according to a reverse
diagonal scan order until quantization unit 106 encounters a
coefficient whose quantized level is greater than 1.
[0156] FIG. 8 is a flowchart illustrating an example portion of
operation 350 of FIG. 7. As indicated in the example of FIG. 8,
quantization unit 106 may determine whether there are any remaining
non-zero coefficients in the coefficient block (400). In response
to determining that there are no remaining non-zero coefficients in
the coefficient block ("NO" of 400), quantization unit 106 may have
determined the optimal values of each coefficient in the
coefficient block and, accordingly, may continue performing the
portion of operation 350 illustrated in FIG. 7.
[0157] On the other hand, in response to determining that there are
one or more remaining non-zero coefficients in the coefficient
block ("YES" of 400), quantization unit 106 may determine whether
the current value of the current coefficient is less than a
reconstructed value of the current coefficient (402). The current
value of the current coefficient may be the value of the current
coefficient prior to quantization. The reconstructed value of the
current coefficient may be equal to Level*StepSize, where Level is
the quantized level for the current coefficient, as determined in
action 354 of FIG. 7. StepSize is the size of the quantization step
4 used to quantize the current coefficient.
[0158] In response to determining that the current value of the
current coefficient is less than the reconstructed value of the
current coefficient ("YES" of 402), quantization unit 106 may
determine whether the quantized level (Level) of the current
coefficient is less than 3 (404). In response to determining that
the quantized level for the current coefficient is less than 3
("YES" of 404), quantization unit 106 may determine the RD cost for
replacing the current coefficient with 0 (406). After determining
the RD cost for replacing the current coefficient with 0 or in
response to determining that the quantized level for the current
coefficient is not less than 3 ("NO" of 404), quantization unit 106
may determine the RD cost for replacing the current coefficient
with the quantized level for the current coefficient (i.e., Level)
(408). In addition, quantization unit 106 may determine the RD cost
for replacing the current coefficient with the quantized level for
the current coefficient (i.e., Level-1) (410). Furthermore,
quantization unit 106 may select an optimal quantized level for the
current coefficient as whichever of Level, Level-1, and 0 has the
lowest RD cost (412). In this way, quantization unit 106 may select
an actual quantized level that produces a lowest rate-distortion
cost.
[0159] After selecting the optimal quantized level for the current
coefficient or in response to determining that the current value of
the current coefficient is not less than the reconstructed value of
the current coefficient ("NO" of 402), quantization unit 106 may
update the RD cost of quantizing all coefficients of the
coefficient block to 0 (414). In addition, quantization unit 106
may update the RD cost of encoding each coefficient of the
coefficient block (416). Quantization unit 106 may update the RD
cost of encoding each coefficient of the coefficient block by
adding the RD cost of the current coefficient to a running sum of
the RD costs of previously-processed coefficients of the
coefficient block. Furthermore, quantization unit 106 may update
the RD cost of encoding the significance map of the coefficient
block (418). Quantization unit 106 may determine the optimal value
of the last significant coefficient of the current block based at
least in part on the RD cost of quantizing all coefficients of the
coefficient block to 0, the RD cost of encoding each coefficient of
the coefficient block, and the RD cost of encoding the significance
map of the coefficient block.
[0160] In addition, quantization unit 106 may update the context
based on the determined optimal quantized level for the current
coefficient (420). Quantization unit 106 may use the context in the
level decision of the next coefficient in the coefficient block.
After updating the context, quantization unit 106 may determine
again whether there are any remaining non-zero coefficients (400).
In this way, quantization unit 106 may determine the optimal values
of each of the coefficients in the coefficient block.
[0161] Thus, in operation 350 of FIGS. 7 and 8, the third step of
the RDOQ technique (i.e., determining the optimal level for a
coefficient) may be simplified as: [0162] 3. Decision of the
optimal level [0163] I. Fix last significant coefficient: the first
coefficient that is quantized to non-zero in reverse diagonal scan
[0164] II. For the remaining non-zero coefficients in the reverse
diagonal scan, choose optimal level based on RD cost [0165] a. If
currentValue<Level*StepSize, that is, [0166]
currentValue<reconstructedValue, Choose the best level
L.sub.Lopt, [0167] i. If Level<3, Calculate the RD cost of
Level, Level-1 and 0 [0168] ii. If Level>=3, Calculate the RD
cost of Level, Level-1 [0169] b. Update encoding cost [0170] i.
Cost of quantized to 0 (for uncoded block) [0171] ii. Cost of
encoding each coefficient [0172] iii. Cost of encoding significance
map [0173] c. Update context based on the decided level L.sub.opt.
The context will be used by the level decision of the next
coefficient [0174] III. Encode the next coefficient, and go to step
II.
[0175] The performance of the example techniques of FIGS. 6-8 is
shown in Tables 6-9, below. As shown in Tables 6-9, there is almost
no RD performance loss incurred by implementing the techniques of
FIGS. 6-8. Furthermore, a level decision saving Level.sub.save may
be calculated. Level.sub.save is defined as follows:
Level save = N [ n .DELTA. , ( n + 1 2 ) .DELTA. ] N total
##EQU00003##
[0176] In the equation above,
N [ n .DELTA. , ( n + 1 2 ) .DELTA. ] ##EQU00004##
is the number of values that fall in the range from n.DELTA. to
(n+1/2).DELTA. and N.sub.total is the total number of values for
which a level decision is performed. Experimental results show that
Level.sub.save is between 60% and 80%. That is, 60% to 80% of the
level decisions can be skipped and the computation involved in RDOQ
has been reduced.
TABLE-US-00006 TABLE 6 All Intra HE Y U V Class A 0.0% 0.0% 0.0%
Class B 0.0% 0.0% 0.0% Class C 0.0% 0.0% 0.0% Class D 0.0% 0.0%
0.0% Class E 0.0% 0.0% 0.0% Class F 0.0% 0.0% 0.0% Overall 0.0%
0.0% 0.0% 0.0% 0.0% 0.0% Enc Time [%] 101%
TABLE-US-00007 TABLE 7 Low delay B HE Y U V Class A Class B 0.1%
0.1% 0.1% Class C 0.1% 0.1% 0.0% Class D 0.0% 0.1% 0.1% Class E
0.0% 1.0% 0.9% Class F -0.1% 0.3% 0.1% Overall 0.0% 0.3% 0.2% 0.0%
0.2% 0.2% Enc Time [%] 99%
TABLE-US-00008 TABLE 8 Random Access HE Y U V Class A -0.1% 1.9%
2.0% Class B 0.0% 0.2% 0.1% Class C 0.0% 0.1% 0.2% Class D 0.0%
0.1% 0.1% Class E Class F 0.0% 0.2% 0.2% Overall 0.0% 0.5% 0.5%
0.3% 0.6% 0.6% Enc Time [%] 101%
TABLE-US-00009 TABLE 9 Low Delay P HE Y U V Class A Class B 0.1%
0.3% 0.1% Class C 0.0% 0.1% 0.1% Class D 0.0% 0.2% 0.2% Class E
0.0% -0.2% -0.1% Class F 0.0% 0.4% 0.0% Overall 0.0% 0.2% 0.1% 0.0%
0.2% 0.0% Enc Time [%] 102%
[0177] FIG. 9 is a flowchart illustrating an example operation 450
to quantize coefficients of a coefficient block, in accordance with
one or more techniques of this disclosure. Video encoder 20 may
perform operation 450 as an alternative to performing operation 300
of FIG. 6 or operation 350 of FIGS. 7 and 8.
[0178] As illustrated in the example of FIG. 9, transform
processing unit 104 may apply one or more transforms to a residual
sample block in order to generate a coefficient block (452). In
addition, transform processing unit 104 may use respective
positions of respective coefficients in a coefficient block to look
up, in a lookup table, respective quantization offsets for the
respective coefficients (454). In some examples, for each
respective position in the coefficient block, an entry in the
lookup table for the respective position is based at least in part
on frequencies with which a RDOQ algorithm selects Level, Level-1,
or 0 as optimal quantized values for coefficients that are located
at the respective position within other test data coefficient
blocks, wherein Level denotes an initial quantized value for a
coefficient.
[0179] Furthermore, quantization unit 106 may determine, based at
least in part on the respective quantization offsets for the one or
more coefficients, respective quantized levels for the respective
coefficients (456). Quantization unit 106 may determine, based at
least in part on costs associated with the quantized levels for the
one or more coefficients, an optimal last significant coefficient
(458). Entropy encoding unit 116 may entropy encode, based at least
in part on the optimal last significant coefficient, a significance
map of the coefficient block (460). In addition, entropy encoding
unit 116 may entropy encode the quantized levels for coefficients
that have positions within the coefficient block that are before
the position within the coefficient block of the optimal last
significant coefficient (462). Furthermore, video encoder 20 may
output a bitstream that includes the entropy encoded quantized
levels (464).
[0180] FIG. 10 is a flowchart illustrating another example
operation 500 to quantize coefficients of a coefficient block, in
accordance with one or more techniques of this disclosure.
Operation 500 may use similar techniques as those in operation 450
of FIG. 9.
[0181] As illustrated in the example of FIG. 10, quantization unit
106 may initialize a context (502). As previously described,
quantization unit 106 may use the context in determining RD costs.
In addition, quantization unit 106 may initialize a current
coefficient indicator (504). In some examples, quantization unit
106 may initialize the current coefficient indicator to indicate
the last coefficient (e.g., the bottom right coefficient) of the
coefficient block. The current coefficient indicator indicates a
coefficient in the coefficient block. This disclosure may refer to
the coefficient indicated by the current coefficient indicator as
the current coefficient.
[0182] Next, quantization unit 106 may, in accordance with the
techniques of this disclosure, use a lookup table to determine a
quantization offset for the current coefficient (506). Thus,
quantization unit 106 may use the position of the current
coefficient to look up a quantization offset for the current
coefficient. In some examples, the lookup table may include a
quantization offset for each position in the coefficient block. In
such examples, for each respective position in the coefficient
block, a quantization offset in the lookup table for the respective
position is based at least in part on frequencies with which
coefficients at the respective position in other coefficient blocks
are quantized to Level, Level-1, and 0, where Level is an initial
quantized level for a coefficient.
[0183] In some examples, quantization unit 106 may, for each
respective coefficient, determine
z = w + f .DELTA. sign ( W ) ( 1 ) ##EQU00005##
where z is the quantized level for the respective coefficient, W is
the initial value of the respective coefficient, f is the
quantization offset for the coefficient, .DELTA. is a quantization
step, [] denotes rounding to the nearest integer towards minus
infinity and sign( ) is a function that returns the sign of the
initial value of the respective coefficient.
[0184] Furthermore, quantization unit 106 may determine, based at
least in part on the determined quantization offset, a quantized
level (Level) for the current coefficient (508). Quantization unit
106 may use equation (1) above to determine the quantized level for
the current coefficient.
[0185] Quantization unit 106 may then update the RD cost of
quantizing all coefficients of the coefficient block to 0 (510). In
addition, quantization unit 106 may update the RD cost of encoding
each coefficient of the coefficient block (512). Furthermore,
quantization unit 106 may update the RD cost of encoding the
significance map of the coefficient block (514). Quantization unit
106 may determine the optimal value of the last significant
coefficient of the current block based at least in part on the RD
cost of quantizing all coefficients of the coefficient block to 0,
the RD cost of encoding each coefficient of the coefficient block,
and the RD cost of encoding the significance map of the coefficient
block.
[0186] Quantization unit 106 may then determine whether there are
any remaining coefficients in the coefficient block (516). In
response to determining that there are one or more remaining
coefficients in the coefficient block ("YES" of 516), quantization
unit 106 may update the current coefficient indicator such that the
current coefficient indicator indicates a next coefficient in the
coefficient block, according to the reverse diagonal scan order
(518). Quantization unit 106 may then use the lookup table again to
determine a quantization offset for the new current coefficient
(506). In this way, quantization unit 160 may determine the
quantized levels for the coefficients in a reverse diagonal scan
order. Furthermore, in this way, quantization unit 106 may, for
each respective remaining coefficient in the coefficient block, use
the lookup table to determine a quantization level for the
respective remaining coefficient and may determine a quantized
level for the respective remaining coefficient based at least in
part on the quantized level for the respective remaining
coefficient.
[0187] On the other hand, in response to determining that there are
no further remaining coefficients in the coefficient block ("NO" of
516), quantization unit 106 may initialize a best code value to be
equal to an RD cost if the current block is un-coded (520). In
addition, quantization unit 106 may initialize a current
coefficient indicator (522). In some examples, quantization unit
106 may initialize the current coefficient indicator to indicate
the last coefficient (e.g., the bottom right coefficient) of the
coefficient block. The current coefficient indicator indicates a
coefficient in the coefficient block. This disclosure may refer to
the coefficient indicated by the current coefficient indicator as
the current coefficient.
[0188] Next, quantization unit 106 may determine whether the
current coefficient has a quantized level greater than 1 (524). In
response to determining that the current coefficient has a
quantized level greater than 1 ("YES" of 524), quantization unit
106 may end operation 500. On the other hand, in response to
determining that the current coefficient does not have a quantized
level greater than 1 ("NO" of 524), quantization unit 106 may
calculate a total block RD cost that would result if the current
coefficient is the last significant coefficient (526).
[0189] Furthermore, quantization unit 106 may determine whether the
calculated total block RD cost is less than the best cost value
(528). In response to determining that the calculated total block
RD cost is less than the best cost value ("YES" of 528),
quantization unit 106 may set a best last position indicator to the
current position and may set the best cost value to the calculated
total block RD cost (530). The current position is the position of
the current coefficient within the coefficient block.
[0190] After setting the best last position indicator and the best
cost value or after determining that the calculated total block RD
cost is not less than the best cost value ("NO" of 528),
quantization unit 106 may update the current coefficient indicator
such that the current coefficient indicator indicates a next
coefficient in the coefficient block, according to the reverse
diagonal scan order (532). Quantization unit 106 may then determine
again whether the current coefficient has a quantized level greater
than 1 (524). In this way, quantization unit 106 may scan through
the coefficients of the coefficient block according to a reverse
diagonal scan order until quantization unit 106 encounters a
coefficient whose quantized level is greater than 1. Furthermore,
in this way, the costs associated with the quantized levels for the
one or more coefficients include rate distortion costs associated
with selecting each one of the one or more coefficients as a last
significant coefficient.
[0191] Thus, the operation of FIG. 10 may implement the following
simplified RDOQ algorithm: [0192] 1. Initialize context: C1=1,
C2=0; [0193] 2. Lookup quantization offset table, quantize the
coefficients with the offset from the quantization table in reverse
diagonal scan order; [0194] 3. Cost calculation for last
significant coefficient decision in step 5 [0195] 1) Cost of
quantized to 0 (for uncoded block) [0196] 2) Cost of encoding each
coefficient [0197] 3) Cost of encoding significance map [0198] 4.
Process the next coefficient, and go to step 2. [0199] 5. Decision
of the optimal last significant coefficient [0200] 1) Initialize
d64BestCost to be the RD cost if the current block is uncoded;
[0201] 2) In reverse diagonal scan, calculate the total block cost
totalCost if the current coefficient is the last significant
coefficient [0202] 3) If totalCost<d64BestCost, new best last
position (iBestLastIdxP1) is current position,
d64BestCost=totalCost [0203] 4) Iterate until
LevelofCurCoeff>1.
[0204] Tables 10-13, below, illustrate example performance of the
operations of FIGS. 9 and 10 with the quantization offset being 1/2
with skipped level decision (i.e., the third step of RDOQ) and
optimal last significant coefficient, as compared to the
performance of a video encoder with RDOQ on. Comparing the
simplified RDOQ algorithm of FIGS. 9 and 10 and the original RDOQ
algorithm, the selection of the RD optimal level now is replaced by
quantization with offset. This may reduce the complexity of RDOQ
with a small RD performance loss. For instance, if video encoder 20
selects the quantization offset of 1/2 and the level decision
(i.e., the third step of RDOQ) is skipped, the performance of the
techniques of FIGS. 9 and 10 compared with RDOQ, as shown in Tables
10-13, shows a RD performance loss from 1% to 3%.
TABLE-US-00010 TABLE 10 All Intra HE Y U V Class A 1.0% 3.8% 3.5%
Class B 1.2% 3.2% 3.3% Class C 1.0% 2.0% 2.1% Class D 1.0% 2.4%
2.0% Class E 1.1% 3.6% 3.1% Class F 0.7% 0.9% 1.1% Overall 1.0%
2.6% 2.5% 1.0% 2.6% 2.5% Enc Time [%] 96% Dec Time [%] #NUM!
TABLE-US-00011 TABLE 11 Low delay B HE Y U V Class A Class B 3.2%
3.2% 2.8% Class C 3.2% 3.1% 2.5% Class D 2.7% 2.8% 1.8% Class E
2.4% 1.5% 2.3% Class F 1.7% 1.7% 2.3% Overall 2.7% 2.6% 2.3% 2.7%
2.5% 2.3% Enc Time [%] 94% Dec Time [%] #NUM!
TABLE-US-00012 TABLE 12 Random Access HE Y U V Class A 2.3% 5.0%
5.2% Class B 1.7% 2.5% 2.5% Class C 1.7% 1.9% 1.9% Class D 1.6%
1.8% 1.2% Class E Class F 1.1% 1.5% 1.3% Overall 1.7% 2.5% 2.4%
1.7% 2.6% 2.5% Enc Time [%] 94% Dec Time [%] #NUM!
TABLE-US-00013 TABLE 13 Low Delay P HE Y U V Class A Class B 3.3%
3.6% 3.4% Class C 3.4% 3.6% 3.1% Class D 3.0% 2.9% 2.8% Class E
2.9% 2.2% 1.6% Class F 2.0% 2.7% 2.5% Overall 3.0% 3.1% 2.8% 3.0%
3.0% 2.7% Enc Time [%] 101% Dec Time [%] #NUM!
[0205] Tables 14-17, below, illustrate the example performance of
the operations of FIGS. 9 and 10 with the quantization offset being
1/3 for intra and 1/6 for inter with skipped level decision and
optimal last significant coefficient, where anchor has RDOQ on.
That is, if the quantization offset is chosen to be 1/3 for intra
mode and 1/6 for inter mode, Tables 14-17 show that the RD
performance loss is from 1.9% to 3.6%.
TABLE-US-00014 TABLE 14 All Intra HE Y U V Class A 2.5% 0.6% 0.5%
Class B 2.5% 1.6% 1.7% Class C 2.1% 1.6% 1.5% Class D 2.1% 1.4%
1.5% Class E 2.4% 1.5% 1.6% Class F 2.7% 1.8% 2.1% Overall 2.4%
1.4% 1.5% 2.4% 1.5% 1.5% Enc Time [%] 91% Dec Time [%] #NUM!
TABLE-US-00015 TABLE 15 Low delay B HE Y U V Class A Class B 1.9%
3.8% 2.9% Class C 1.9% 4.1% 4.0% Class D 2.4% 5.9% 5.3% Class E
0.7% 3.5% 2.8% Class F 3.4% 3.6% 4.0% Overall 2.1% 4.2% 3.8% 2.1%
4.1% 3.7% Enc Time [%] 90% Dec Time [%] #NUM!
TABLE-US-00016 TABLE 16 Random Access HE Y U V Class A 3.3% 6.4%
7.3% Class B 3.7% 6.8% 6.3% Class C 3.6% 7.1% 7.1% Class D 3.8%
6.7% 6.2% Class E Class F 3.9% 5.6% 5.3% Overall 3.6% 6.5% 6.4%
3.9% 6.5% 6.4% Enc Time [%] 91% Dec Time [%] #NUM!
TABLE-US-00017 TABLE 17 Low Delay P HE Y U V Class A Class B 1.4%
3.3% 2.4% Class C 1.7% 4.0% 4.1% Class D 2.3% 5.7% 5.3% Class E
0.4% 2.7% 2.5% Class F 3.6% 3.3% 4.0% Overall 1.9% 3.8% 3.6% 1.8%
3.8% 3.5% Enc Time [%] 92% Dec Time [%] #NUM!
[0206] Tables 18-21, below, illustrate example performance of the
operations of FIGS. 9 and 10 with the quantization offset being 1/2
for intra and 1/6 for inter with skipped level decision and optimal
last significant coefficient, where anchor has RDOQ on. That is, if
the quantization offset is set to be 1/2 for intra mode and 1/6 for
inter mode, the RD performance loss may be from 1% to 2.5%.
TABLE-US-00018 TABLE 18 All Intra HE Y U V Class A 1.0% 3.8% 3.5%
Class B 1.2% 3.2% 3.3% Class C 1.0% 2.0% 2.1% Class D 1.0% 2.4%
2.0% Class E 1.1% 3.6% 3.1% Class F 0.7% 0.9% 1.1% Overall 1.0%
2.6% 2.5% 1.0% 2.6% 2.5% Enc Time [%] 100% Dec Time [%] 101%
TABLE-US-00019 TABLE 19 Low delay B HE Y U V Class A Class B 1.7%
3.6% 2.6% Class C 1.8% 4.1% 3.6% Class D 2.2% 5.4% 5.3% Class E
0.1% 2.1% 1.5% Class F 2.4% 1.8% 2.3% Overall 1.7% 3.5% 3.1% 1.7%
3.4% 3.0% Enc Time [%] 97% Dec Time [%] 98%
TABLE-US-00020 TABLE 20 Random Access HE Y U V Class A 2.7% 4.5%
4.2% Class B 2.6% 4.8% 3.8% Class C 2.6% 5.3% 5.3% Class D 2.8%
4.6% 4.1% Class E Class F 1.6% 2.5% 2.1% Overall 2.5% 4.3% 3.9%
2.8% 4.4% 3.9% Enc Time [%] 97% Dec Time [%] 99%
TABLE-US-00021 TABLE 21 Low Delay P HE Y U V Class A Class B 1.2%
3.0% 2.2% Class C 1.6% 3.7% 3.8% Class D 2.3% 5.2% 4.9% Class E
-0.2% 2.0% 0.7% Class F 2.7% 2.4% 2.2% Overall 1.6% 3.3% 2.9% 1.5%
3.3% 2.7% Enc Time [%] 98% Dec Time [%] 99%
[0207] FIG. 11 is a flowchart illustrating an example encoding
operation 650, in accordance with one or more techniques of this
disclosure. FIG. 11 is primarily for the purpose of illustrating
the modules involved in RDOQ. Consequently, some processes are
omitted from the example of FIG. 11 and simplified as blocks such
as "intermediate processing" and "other processing." Some video
encoders may, after performing some intermediate processing such as
CU/PU split, mode decision, inter/intra prediction, etc., for every
TU, perform bit estimation by calculating the entropy of each
context model. The conventional video encoder may then perform RDOQ
using the estimated bits. Consequently, such video encoders may
perform bit estimation for every TU and during RDOQ the cost of
rate (.lamda.R) may need to be calculated all the time.
[0208] In contrast, video encoder 20 may, when performing operation
650 of FIG. 11, perform bit estimation only once at the beginning
of encoding a picture. Video encoder 20 may then generate a
multiplication table (for .lamda.R) for each of the context models.
Later, when video encoder 20 performs the RDOQ algorithm, video
encoder 20 may look up a value in the multiplication table to
calculate the RD cost.
[0209] In the example of FIG. 11, video encoder 20 may determine
whether the current frame is the last frame (652). In response to
determining that the current frame is not the last frame ("NO" of
652), video encoder 20 may perform a bit estimation operation
(654). When video encoder 20 performs the bit estimation operation,
video encoder 20 may determine, based at least in part on the
entropies of the context models used in the previous frame, the
entropies of each context model used in the current frame. For each
respective context model used in the current frame, video encoder
20 may use the entropy of the respective context model to determine
a value of R.sub.level, a value of R.sub.sig.sub.--.sub.map, a
value of R.sub.positionX, and a value of R.sub.positionY for each
applicable quantized level, significance map flag, last significant
coefficient (LSC) position X and LSC position Y, respectively. As
described above, R.sub.level is the number of bits estimated to
encode a quantized level. R.sub.sig.sub.--.sub.map is the number of
bits estimated to encode a significance map of a coefficient.
R.sub.positionX and R.sub.positionY are the number of bits
estimated to encode the position of the last significant
coefficient.
[0210] After performing the bit estimation operation, video encoder
20 may generate a multiplication table (656). The multiplication
table may specify a value of .lamda.R.sub.level for each
combination of a context and a quantized level value. In addition,
the multiplication table may specify a value of
.lamda.R.sub.sig.sub.--.sub.map for each combination of a context
and a significance map flag value. The multiplication table may
also specify a value of .lamda.R.sub.positionX for each combination
of a context and an X coordinate of a LSC. The multiplication table
may also specify a value of .lamda.R.sub.positionY for each
combination of a context and a Y coordinate of a LSC.
[0211] Following generation of the multiplication table, video
encoder 20 may perform intermediate processing (658). The
intermediate processing may include partitioning the current frame
into CUs, partitioning CUs into PUs and TUs, selecting inter or
intra prediction modes for the PUs, and so on.
[0212] Video encoder 20 may then determine whether every TU of the
current frame has been processed (660). In response to determining
that not every TU of the current frame has been processed ("NO" of
660), video encoder 20 may perform TU processing for a current TU
(662). When video encoder 20 performs TU processing for the current
TU, video encoder 20 may apply one or more transforms to the
residual sample blocks of the TU.
[0213] Video encoder 20 may then perform RDOQ on the coefficient
blocks of the current TU (664). When video encoder 20 performs RDOQ
on a coefficient block of the current TU, video encoder 20 may
determine an optimal quantized value for each coefficient of the
coefficient block. When determining an optimal quantized value for
a particular coefficient, video encoder 20 may use the
multiplication table to look up a value of .lamda.R.sub.level for a
current quantized level context and a quantized level for the
particular coefficient. Furthermore, when determining the optimal
quantized value for the particular coefficient, video encoder 20
may use the multiplication table to look up a value of
.lamda.R.sub.sig.sub.--.sub.map for a current significance map flag
context and a value of the significance map flag for the particular
coefficient. In addition, when determining the optimal last
significant coefficient, video encoder 20 may use the
multiplication table to look up a value of .lamda.R.sub.positionX
for a current LSC x-coordinate context and an LSC x-coordinate of
the LSC. When determining the optimal last significant coefficient,
video encoder 20 may use the multiplication table to look up a
value of .lamda.R.sub.positionY for a current LSC y-coordinate
context and an LSC y-coordinate of the LSC. The dashed arrow in the
example of FIG. 11 indicates the use of the multiplication table
during RDOQ. By looking up values of .lamda.R.sub.level,
.lamda.R.sub.sig.sub.--.sub.map, .lamda.R.sub.positionX, and
.lamda.R.sub.positionY when performing RDOQ instead of performing
double point multiplications to determine the values of
.lamda.R.sub.level, .lamda.R.sub.sig.sub.--.sub.map,
.lamda.R.sub.positionX, and .lamda.R.sub.positionY, the computation
time of performing RDOQ may be decreased.
[0214] After performing RDOQ on the coefficient blocks of the
current TU, video encoder 20 may perform any remaining TU
processing (666). The remaining TU processing may include CABAC
encoding syntax elements that indicate quantized values of
coefficients of the current TU. Video encoder 20 may then determine
again whether every TU of the current frame has been processed
(660). In response to determining that every TU has been processed
for the current frame ("YES" of 660), video encoder 20 may perform
other processing for the current frame (668). The other processing
for the current frame may include loop filtering, bitstream
generation, and generating the reconstructed frame for next frame's
prediction, and so on. After performing the other processing for
the current frame, video encoder 20 may determine again whether the
current frame is the last frame (652). If the current frame is not
the last frame ("NO" of 652), video encoder 20 may perform actions
654-668 with regard to a next frame. Otherwise, in response to
determining that the current frame is the last frame ("YES" of
652), video encoder 20 may end operation 650.
[0215] In this way, quantization unit 106 may determine, based at
least in part on entropies of context models of a previous frame,
rate values for the context models. Furthermore, quantization unit
160 may generate a table that indicates, for each of the context
models, a value of X multiplied by the rate value for the context
model, where .lamda. is a fixed value. Determining the
rate-distortion costs of quantizing the coefficient to be the
initial quantized level for the coefficient, the initial quantized
level minus one, and zero may comprise determining, based at least
in part on the values in the table, the rate-distortion costs of
quantizing the coefficient to be the initial quantized level for
the coefficient, the initial quantized level minus one, and
zero
[0216] FIG. 12 is a flowchart illustrating an example coding
operation 700, in accordance with the techniques of this
disclosure. Video encoder 20 may operate in various coding modes,
such as All Intra (AI) mode, random access (RA) mode, low delay B
(LB) mode and low delay P (LP) mode. In the All Intra configuration
(i.e., the intra-only configuration) each picture in a video
sequence is encoded as an instantaneous decoding refresh (IDR)
picture. An IDR picture contains only I slices.
[0217] In the RA configuration, intra pictures are inserted into
the bitstream cyclically. Furthermore, in the RA configuration, a
first intra picture of a video sequence is encoded as an IDR
picture and the other intra pictures are encoded as non-IDR
pictures. In the RA configuration, pictures located between
successive intra pictures in the display order are encoded as B
pictures. The generalized P and B (GPB) picture is used as the
lowest temporal layer that can refer to I or GPB picture for inter
prediction. The second and third temporal layers may consist of
referenced B pictures, and the highest temporal layer contains
non-referenced B picture only. The QP of each inter coded picture
may be derived by adding offset to the QP of an intra coded picture
depending on the temporal layer. A reference picture list
combination may be used for management and entropy coding of
reference picture index.
[0218] In the low delay configurations (i.e., the low delay B
configuration and the low delay P configuration), only the first
picture in a video sequence is encoded as IDR picture. In the
low-delay B configuration, the other successive pictures shall be
encoded as B pictures. In the low-delay P configuration, the other
successive pictures shall be encoded as P pictures. The B or P
pictures (i.e., generalized P or B (GPB) pictures) use only the
reference pictures, each of whose picture order count (POC) values
is smaller than that of the current picture (i.e., all reference
pictures in RefPicList0 and RefPicList1 shall be temporally
previous in display order relative to the current picture). In the
low delay B and P configurations, the contents of RefPicList0 and
RefPicList1 are identical.
[0219] In the example of FIG. 12, a video coder (such as video
encoder 20 or video decoder 30) may determine, based at least in
part on whether the video encoder is operating in an all-intra
configuration, a random access configuration, or a low-delay
configuration, a quantization offset (702). Furthermore, the video
coder may determine, based at least in part on the quantization
offset, respective quantized levels for the coefficients (704).
[0220] In some examples, the quantization offset is equal to 1/2
when the video encoder is operating in the all-intra configuration
or the random access configuration. In such examples, the
quantization offset is equal to 1/2 when the video encoder is
operating in the low-delay configuration and the coefficients are
associated with an intra slice. Furthermore, in such examples, the
quantization offset is equal to 1/6 when the video encoder is
operating in the low-delay configuration and the coefficients are
associated with an inter slice.
[0221] Table 22, below, illustrates an example performance
comparison for different offsets and RDOQ off at different
configurations of HEVC, such as All Intra (AI), random access (RA),
low delay B (LB) and low delay P (LP). That is, Table 22, below,
compares the performances of different quantization offsets and
RDOQ off. From Table 22 another way of generating the quantization
offset is obtained. For AI and RA configurations, a quantization
offset 1/2 is used as it may give the best performance. For LB and
LP configuration, intra slice should use the quantization offset
1/2 and inter slice should use the quantization offset 1/6 because
it gives the best performance. Tables 10-22 show that by choosing
an appropriate quantization offset, RD performance can approximate
that of RDOQ with reduced complexity.
TABLE-US-00022 TABLE 22 Offset 1/3 Offset 1/2 (intra) and 1/6
(intra) and 1/6 Offset 1/2 (inter) (inter) RDOQ off AI 1% 2.4% 1%
5.7% RA 1.7% 3.6% 2.5% 5.3% LB 2.7% 2.1% 1.7% 3.5% LP 3% 1.9% 1.6%
3.4%
[0222] In one or more examples, the functions described may be
implemented in hardware, software, firmware, or any combination
thereof. If implemented in software, the functions may be stored on
or transmitted over, as one or more instructions or code, a
computer-readable medium and executed by a hardware-based
processing unit. Computer-readable media may include
computer-readable storage media, which corresponds to a tangible
medium such as data storage media, or communication media including
any medium that facilitates transfer of a computer program from one
place to another, e.g., according to a communication protocol. In
this manner, computer-readable media generally may correspond to
(1) tangible computer-readable storage media which is
non-transitory or (2) a communication medium such as a signal or
carrier wave. Data storage media may be any available media that
can be accessed by one or more computers or one or more processors
to retrieve instructions, code and/or data structures for
implementation of the techniques described in this disclosure. A
computer program product may include a computer-readable
medium.
[0223] By way of example, and not limitation, such
computer-readable storage media can comprise RAM, ROM, EEPROM,
CD-ROM or other optical disk storage, magnetic disk storage, or
other magnetic storage devices, flash memory, or any other medium
that can be used to store desired program code in the form of
instructions or data structures and that can be accessed by a
computer. Also, any connection is properly termed a
computer-readable medium. For example, if instructions are
transmitted from a website, server, or other remote source using a
coaxial cable, fiber optic cable, twisted pair, digital subscriber
line (DSL), or wireless technologies such as infrared, radio, and
microwave, then the coaxial cable, fiber optic cable, twisted pair,
DSL, or wireless technologies such as infrared, radio, and
microwave are included in the definition of medium. It should be
understood, however, that computer-readable storage media and data
storage media do not include connections, carrier waves, signals,
or other transient media, but are instead directed to
non-transient, tangible storage media. Disk and disc, as used
herein, includes compact disc (CD), laser disc, optical disc,
digital versatile disc (DVD), floppy disk and Blu-ray disc, where
disks usually reproduce data magnetically, while discs reproduce
data optically with lasers. Combinations of the above should also
be included within the scope of computer-readable media.
[0224] Instructions may be executed by one or more processors, such
as one or more digital signal processors (DSPs), general purpose
microprocessors, application specific integrated circuits (ASICs),
field programmable logic arrays (FPGAs), or other equivalent
integrated or discrete logic circuitry. Accordingly, the term
"processor," as used herein may refer to any of the foregoing
structure or any other structure suitable for implementation of the
techniques described herein. In addition, in some aspects, the
functionality described herein may be provided within dedicated
hardware and/or software modules configured for encoding and
decoding, or incorporated in a combined codec. Also, the techniques
could be fully implemented in one or more circuits or logic
elements.
[0225] The techniques of this disclosure may be implemented in a
wide variety of devices or apparatuses, including a wireless
handset, an integrated circuit (IC) or a set of ICs (e.g., a chip
set). Various components, modules, or units are described in this
disclosure to emphasize functional aspects of devices configured to
perform the disclosed techniques, but do not necessarily require
realization by different hardware units. Rather, as described
above, various units may be combined in a codec hardware unit or
provided by a collection of interoperative hardware units,
including one or more processors as described above, in conjunction
with suitable software and/or firmware.
[0226] Various examples have been described. These and other
examples are within the scope of the following claims.
* * * * *
References