U.S. patent application number 12/512628 was filed with the patent office on 2010-11-04 for method for image recombination of a plurality of images and image identification and system for image acquiring and identification.
This patent application is currently assigned to INDUSTRIAL TECHNOLOGY RESEARCH INSTITUTE. Invention is credited to Chun-Lung Chang, Lai-Sheng Chen, Kuo-Tang Huang, Ya-Hui Tsai.
Application Number | 20100278389 12/512628 |
Document ID | / |
Family ID | 43030372 |
Filed Date | 2010-11-04 |
United States Patent
Application |
20100278389 |
Kind Code |
A1 |
Tsai; Ya-Hui ; et
al. |
November 4, 2010 |
METHOD FOR IMAGE RECOMBINATION OF A PLURALITY OF IMAGES AND IMAGE
IDENTIFICATION AND SYSTEM FOR IMAGE ACQUIRING AND
IDENTIFICATION
Abstract
The present invention provides a method for image recombination
of a plurality of images and image identification and a system for
image acquiring and identification. Features with respect to the
plurality of images are recombined and enhanced so as to form a
recombined image. After that, the recombined image is processed to
emphasize the features of the recombined image so that the
recombined image is capable of being identified easily.
Furthermore, the present provides a system to perform the foregoing
method, whereby reducing unidentified problems caused due to low
quality image of the monitoring system.
Inventors: |
Tsai; Ya-Hui; (Taoyuan
County, TW) ; Huang; Kuo-Tang; (Taoyuan County,
TW) ; Chang; Chun-Lung; (Hsinchu City, TW) ;
Chen; Lai-Sheng; (Hsinchu City, TW) |
Correspondence
Address: |
RABIN & Berdo, PC
1101 14TH STREET, NW, SUITE 500
WASHINGTON
DC
20005
US
|
Assignee: |
INDUSTRIAL TECHNOLOGY RESEARCH
INSTITUTE
Hsinchu
TW
|
Family ID: |
43030372 |
Appl. No.: |
12/512628 |
Filed: |
July 30, 2009 |
Current U.S.
Class: |
382/105 ;
382/190; 382/284 |
Current CPC
Class: |
G06K 2209/01 20130101;
G06K 2209/15 20130101; G06K 9/325 20130101; G06K 9/4604 20130101;
G06K 9/3258 20130101; G06K 9/72 20130101; G06T 11/60 20130101 |
Class at
Publication: |
382/105 ;
382/190; 382/284 |
International
Class: |
G06K 9/00 20060101
G06K009/00 |
Foreign Application Data
Date |
Code |
Application Number |
Apr 30, 2009 |
TW |
098114328 |
Claims
1. A method for image recombination of a plurality of images,
comprising steps of: acquiring a plurality of images; determining a
region of interest in an image from the plurality of images and
acquiring features in the region of interest; acquiring from other
regions a feature region corresponding to the region of interest
according to the acquired features; and performing an image
recombination process to form a recombined image according to a
plurality of feature regions and the region of interest.
2. The method for image recombination of a plurality of images as
recited in claim 1, further comprising a step of performing an
image enhancing process on the recombined image.
3. The method for image recombination of a plurality of images as
recited in claim 1, wherein the image recombination process
comprises steps of: performing an operation process on
corresponding pixels in the plurality of feature regions and the
region of interest; and forming the recombined image after the
operation process.
4. The method for image recombination of a plurality of images as
recited in claim 3, wherein the corresponding pixels are averaged
in the operation process.
5. The method for image recombination of a plurality of images as
recited in claim 4, further comprising a step of performing a
histogram equalization process on the recombined image.
6. The method for image recombination of a plurality of images as
recited in claim 3, wherein the operation process further comprises
steps of: performing an inversing operation on the pixels in the
plurality of feature regions and the pixels in the region of
interest respectively; summing the pixels in the plurality of
feature regions and the pixels in the region of interest to form
the recombined image; and performing an inversing operation on the
recombined image.
7. The method for image recombination of a plurality of images as
recited in claim 1, wherein the region of interest is determined by
choosing from the plurality of images an image as a standard image
and choosing from the standard image a region as the region of
interest.
8. The method for image recombination of a plurality of images as
recited in claim 1, further comprising a step of normalizing the
plurality of images.
9. A method for image identification, comprising steps of:
acquiring a plurality of images; determining a region of interest
in an image from the plurality of images and acquiring features in
the region of interest; acquiring from other regions a feature
region corresponding to the region of interest according to the
acquired features; and performing an image recombination process to
form a recombined image according to a plurality of feature regions
and the region of interest; and identifying the recombined
image.
10. The method for image identification as recited in claim 9,
wherein the step of identifying the recombined image further
comprises steps of: acquiring at least one feature image in the
recombined image; providing a database comprising a plurality of
sample images therein, each having respectively a standard image
region and at least a non-standard image region, wherein the
standard image region has pixels corresponding to a first feature
value respectively, and the non-standard image region has pixels
corresponding to a second feature value respectively; performing a
calculation on a third feature value of each pixel in the feature
image and the first feature value or the second feature value
corresponding to each pixel in the plurality of sample images to
obtain a similarity index of the feature image corresponding to the
plurality of sample images respectively; collecting a plurality of
similarity indexes with respect to the feature image compared with
the plurality of sample images; and sorting the plurality of
similarity indexes and outputting at least one of comparison
results.
11. The method for image identification as recited in claim 10,
wherein the calculation is based on normalized correlation
matching.
12. The method for image identification as recited in claim 10,
further comprising a step of normalizing the feature image to
adjust the size and the angle of the feature image so that the size
of the feature image is identical to the size of the sample image
after acquiring the feature image from the image.
13. The method for image identification as recited in claim 10,
wherein each sample image corresponds to an image of a digit or a
character.
14. The method for image identification as recited in claim 10,
wherein the first feature value and the second feature value are
respectively a combination of a weight value and a greyscale value,
and the third feature value is a greyscale value.
15. The method for image identification as recited in claim 9,
wherein the recombined image comprises information regarding an
identification mark on a carrier.
16. The method for image identification as recited in claim 9,
wherein the image recombination process comprises steps of:
performing an operation process on corresponding pixels in a
plurality of feature regions and the region of interest; and
forming the recombined image after the operation process.
17. The method for image identification as recited in claim 16,
wherein the corresponding pixels are averaged in the operation
process.
18. The method for image identification as recited in claim 17,
further comprising a step of performing a histogram equalization
process on the recombined image.
19. The method for image identification as recited in claim 16,
wherein the operation process further comprises steps of:
performing an inversing operation on the pixels in the plurality of
feature regions and the pixels in the region of interest
respectively; summing the pixels in the plurality of feature
regions and the pixels in the region of interest to form the
recombined image; and performing an inversing operation on the
recombined image.
20. A system for image acquiring and identification, comprising: an
image input unit capable of providing a plurality of images; an
image processing unit coupled to the image input unit, the image
processing unit further comprising: a feature acquiring unit
capable of acquiring features in a region of interest on an image
from the plurality of images and acquiring from other regions a
feature region corresponding to the region of interest according to
the acquired features in other regions; a recombination unit
capable of performing an image recombination process according to
the plurality of feature regions and the region of interest to form
a recombined image.
21. The system for image acquiring and identification as recited in
claim 20, wherein the image processing unit further comprises: a
database comprising a plurality of sample images therein, each
having respectively a standard image region and at least a
non-standard image region, wherein the standard image region has
pixels corresponding to a first feature value respectively, and the
non-standard image region has pixels corresponding to a second
feature value respectively; and an identification and comparison
unit coupled to the recombination unit to identify the recombined
image, the identification and comparison unit performing a
calculation on each pixel in the feature image and each pixel in
the plurality of sample images to obtain a similarity index of the
feature image corresponding to the plurality of sample images
respectively, the identification and comparison unit further
collecting a plurality of similarity indexes with respect to the
feature image compared with the plurality of sample images.
22. The system for image acquiring and identification as recited in
claim 20, wherein the image processing unit is further connected to
an output unit to sort the plurality of similarity indexes and
output at least one of comparison results.
23. The system for image acquiring and identification as recited in
claim 20, wherein the recombination unit is further connected to an
enhancing unit to enhance the recombined image.
24. The system for image acquiring and identification as recited in
claim 20, wherein the recombination unit is capable of normalizing
the plurality of images for an image recombination process.
Description
FIELD OF THE INVENTION
[0001] The present invention generally relates to an image
identification technology and, more particularly, to a method for
image recombination of a plurality of images and image
identification and a system for image acquiring and identification,
wherein features with respect to the plurality of images are
recombined so as to identify the content of an image.
BACKGROUND OF THE INVENTION
[0002] There are many people who get killed in traffic accidents.
Theft and burglary using cars/motorcycles have been repeatedly
reported. These may be attributed to poor image identification of
license plates because of poor monitoring systems. Such monitoring
systems are mostly problematic because of poor resolution
(320.times.240 Pixels) and slant angles of the image acquiring
units to cause blur or imcomplete images that cannot be recognized
so that the criminals can be at large.
[0003] Conventionally, in Taiwan Patent No. 197752, a CCD camera
and an image acquiring unit are used to acquire a car image in the
car lane and the car image is then read by an image reading unit.
Then, a logarithmic greyscale operation unit is used to calculate
the logarithmic greyscale of each pixel in the car image. The image
corresponding to the logarithmic greyscales is decomposed by a
wavelet decomposition unit into rough images, horizontally
differentiated images, vertically differentiated images and
diagonally differentiated images. An image binarization unit
converts the logarithmic greyscale of each pixel in the
horizontally differentiated images from real numbers into binary
digits 0 and 1. A rough image dividing unit determines a region
with the highest sum of binary digits within the whole car image
according a pre-set license plate size and thus the region is
initially referred to as a license plate region. Then, a license
plate slantness correction unit corrects the slantness of the image
corresponding to the license plate region. Finally, a fine image
dividing unit removes the part that does not correspond to the
license plate from the rough license plate region.
[0004] Moreover, in Taiwan Patent Pub. No. I286027, an integrated
plurality of lane free flow vehicle enforcement system is
disclosed, wherein a portal framed equipment is established at the
image enforcement point. The car lane is physically divided so that
image enforcement can be realized with respect to various cars even
though the system slows the cars to pass by the image enforcement
point at a normal speed and to change lanes freely.
[0005] Moreover, in Taiwan Patent Appl. No. 200802137, a serial
license plate identification system is disclosed, using a license
plate character region detection module to receive an image and
determine each approximate license plate range in the image.
Sequences of serial identical pixels in each approximate license
plate range are obtained. The sequences of serial identical pixels
are erased, filtered, and connected to blocks so as to obtain the
image with respect to the license plate character region in each
approximate license plate range and output verified image with
respect to the license plate character region after verification.
Then, the verified image with respect to the license plate
character region is transmitted to the a license plate character
dividing and identification module to acquire all the independent
character images and thus all the license plate character
information after the independent character images are
identified.
[0006] Moreover, Taiwan Patent No. 221193 discloses a license plate
identification and monitoring apparatus used in a parking area.
When a car passes by a predetermined image acquiring spot, the host
is informed to enable the duplex image acquiring device to control
the camera device to acquire the car license plate image, which is
then processed by the an identification process to identify the
characters on the license plate for car management, stolen car
seeking and prevention in the parking area.
[0007] Taiwan Patent No. 226454 discloses a license plate
identification method, wherein the logic relation and character
strokes are used to determine the correct license plate location in
the digital image. Then, ternarized difference and fuzzy inference
are used to acquire the outlines of the characters on the license
plate. Adaptive binarization method is used to divide the
boundaries of each character. Finally, the identification result
can be obtained by a feature fused median calculation using a
neural network.
[0008] Moreover, Taiwan Patent No. 191905 discloses a automatic
mobile license plate identification system, which comprises an
image acquiring device and an image processing device that can be
installed in a car to perform automatic identification on a static
or moving car being monitored. The image acquiring device is
capable of acquiring the image of the license plate and
transmitting the image into the image processing device. The image
processing device performs a precise acquiring process on the
license plate characters based on fuzzy inference and performs a
character identification process on the characters using character
structure analysis. Therefore, identification errors due to license
plate contamination, bending, character contamination or deflexion
can be prevented.
[0009] Taiwan Patent No. 123259 discloses a license plate number
identification apparatus installed at a spot where cars pass by so
as to automatically identify the license plate number of a car. The
license plate number identification apparatus uses an image
acquiring device capable of acquiring an image containing the
license plate and an image processing unit capable of checking the
digital image according to features of the license plate number to
find the license plate location, specify the range of characters,
divide the characters to achieve feature identification of each
characters.
[0010] Moreover, U.S. Pat. No. 4,817,166 discloses a method for
reading a license plate by acquiring the boundary features such as
length, height, and width of the characters on the license plate.
With such information regarding the character features, geometric
features of the characters such as the locations and shapes of
convex hulls, turns and holes are analyzed. Finally, the structure
of each character on the license plate is analyzed according to the
results of the aforegoing analysis.
[0011] Moreover, U.S. Pat. No. 6,553,131 discloses an
identification technology using a smart image acquiring device to
perform license plate identification. A processor is installed
inside the image acquiring device to perform license plate
information identification. In this technology, image
identification is implemented by determining a base line according
to the brightness and location of the license plate image and a
blur region. The image having a base line is then processed by
projection to obtain the location of each character on the license
plate. A statistic-based method is used so that each character is
provided with a confidence index. Finally, character information on
the license plate is determined according to the confidence
index.
[0012] Moreover, U.S. Pat. No. 5,425,108 discloses a license plate
image identification technology, wherein the acquired license plate
image is processed by fuzzy interfere and the features of the
license plate image is identified by structure analysis using a
neural network features.
[0013] Moreover, U.S. Pat. No. 6,473,517 discloses a license plate
identification technology, wherein the license plate image is
identified by character segmentation. In this technology, the
license plate image is divided into a plurality of regions to be
converted into possible character regions (or suspected character
regions). Then, the possible character regions are identified to
obtain a confidence index for image identification based
thereon.
[0014] U.S. Pat. No. 5,081,685 discloses a license plate
identification technology, wherein image intensity information is
used to identify the characters on the license plate. In this
technology, the characters are separated from the background on the
license plate so as to obtain the outlines of the characters by a
tracking process.
SUMMARY OF THE INVENTION
[0015] The present invention provides a method for image
recombination of a plurality of images and image identification and
a system for image acquiring and identification, wherein a
plurality of images corresponding to a specific target are
recombined to compensate the incomplete information of each image
to form a recombined image with enhanced image identification.
[0016] The present invention provides a method for image
recombination of a plurality of images and image identification and
a system for image acquiring and identification, wherein a
recombined image formed by recombining a plurality of images is
identified with a plurality of similarity indexes according to the
provided information in a database. The possible results are
obtained according to the similarity indexes for the user to choose
from with enhanced identification speed and precision.
[0017] The present invention provides a method for image
recombination of a plurality of images and image identification and
a system for image acquiring and identification, which is useful in
identification on the identification mark of a carrier. With
enhanced character features of the identification mark, multi-angle
license plate identification technology can be used to help the
user to identify cars that are suspected to cause accidents.
[0018] In one embodiment, the present invention provides a method
for image recombination of a plurality of images, comprising steps
of: acquiring a plurality of images; determining a region of
interest in an image from the plurality of images and acquiring
features in the region of interest; acquiring from other regions a
feature region corresponding to the region of interest according to
the acquired features; and performing an image recombination
process to form a recombined image according to a plurality of
feature regions and the region of interest.
[0019] In another embodiment, the present invention provides a
method for image identification, comprising steps of: acquiring a
plurality of images; determining a region of interest in an image
from the plurality of images and acquiring features in the region
of interest; acquiring from other regions a feature region
corresponding to the region of interest according to the acquired
features; and performing an image recombination process to form a
recombined image according to a plurality of feature regions and
the region of interest; and identifying the recombined image.
[0020] In another embodiment, the present invention provides a
system for image acquiring and identification, comprising: an image
input unit capable of providing a plurality of images; an image
processing unit coupled to the image input unit, the image
processing unit further comprising: a feature acquiring unit
capable of acquiring features in a region of interest on an image
from the plurality of images and acquiring from other regions a
feature region corresponding to the region of interest according to
the acquired features in other regions; a recombination unit
capable of performing an image recombination process according to
the plurality of feature regions and the region of interest to form
a recombined image.
BRIEF DESCRIPTION OF THE DRAWINGS
[0021] The objects and spirits of the embodiments of the present
invention will be readily understood by the accompanying drawings
and detailed descriptions, wherein:
[0022] FIG. 1 is a flowchart of a method for image recombination of
a plurality of images according to one embodiment of the present
invention;
[0023] FIG. 2A and FIG. 2B are schematic diagrams showing a car
that is moving;
[0024] FIG. 3A and FIG. 3B are schematic diagrams showing the
acquired images of a car at different locations;
[0025] FIG. 3C is a schematic diagram showing the angular relation
with respect to the feature region and the region of interest;
[0026] FIG. 4A and FIG. 4B are schematic diagrams of a plurality of
images and a recombined image respectively;
[0027] FIG. 5 is a flowchart of a method for image recombination of
a plurality of images according to another embodiment of the
present invention;
[0028] FIG. 6A and FIG. 6B are schematic diagrams of a plurality of
images and a recombined image respectively;
[0029] FIG. 7A and FIG. 7B show the greyscale value with respect to
the location before and after histogram equalization
respectively;
[0030] FIG. 8 is a flowchart of a method for image identification
according to the present invention;
[0031] FIG. 9A to FIG. 9D are schematic diagrams showing the
formation of a sample image;
[0032] FIG. 10A is a schematic diagram of a recombined image and a
feature image thereof;
[0033] FIG. 10B is a schematic diagram of a feature image;
[0034] FIG. 11 is a table for sorting the comparison results of an
identification mark according to the present invention; and
[0035] FIG. 12 is a schematic diagram of a system for image
acquiring and identification according to the present
invention.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT
[0036] The present invention can be exemplified but not limited by
various embodiments as described hereinafter.
[0037] Please refer to FIG. 1, which is a flowchart of a method for
image recombination of a plurality of images according to one
embodiment of the present invention. In the present embodiment, the
method 2 starts with step 200 to acquiring a plurality of images.
In the present step, the plurality of images can be acquired in
many ways, for example, inputting a plurality of images or using a
camera, a CCD or a CMOS image acquiring unit, but not limited
thereto, to acquire the plurality of images at different timings or
using an image acquiring device to acquire images with sequential
reation from continuous images by a camera. In FIG. 2A, when a car
90 is making a turn, images of the car 90 are acquired at different
timings by an image acquiring unit 25 to acquire a plurality of
images. In FIG. 2B, images at different locations are acquired when
the car 90 is moving. For example, the image at location 91 is as
shown in FIG. 3A, while the image at location 92 is as shown in
FIG. 3B. The number of the images is based on the practical demand,
and is thus not limited. In step 200, the images are images of a
car, but not limited thereto.
[0038] Referring to FIG. 1, after the images are acquired, step 201
is performed to load the plurality of images. Then, in step 212,
the images are stored in a storage medium, such as a hard disk or
memory. Then step 202 is performed to choose from the plurality of
images an image as a standard image. The chosen image can be the
clearest one to be the standard image. In the embodiment, the
standard image is as shown in FIG. 3B. Then, step 203 is performed
to determine a region of interest (ROI) in the standard image. The
region of interest covers the license plate image of a car, as
shown in region 93 in FIG. 3B. Then, step 204 is performed to
acquire the image features in the region of interest. In the
present embodiment, the image features include contrast or
greyscale value of the outlines of the word or digit on the license
plate image in the feature region.
[0039] Then, step 205 is performed to acquire from other regions a
feature region corresponding to the ROI according to the features
acquired in step 204. The feature region is acquired manually or
automatically with the use of software. The feature region is
determined according to the object to be identified. In the present
embodiment, an identification mark of a car is used as an example.
The feature region refers to a region corresponding to the
identification mark. Step 205 comprises two steps as described
herein. Firstly, the image stored in step 212 is read. Then, the
features acquired in step 204 are loaded to perform a feature
searching process on the loaded images. For example, a feature
region 94 can be searched from the image in FIG. 3A according to
features acquired in region 91 in FIG. 3B. In step 205, in addition
to acquiring the feature region, the angular relation and the scale
relation between the feature region and the region of interest are
also determined. For example, in FIG. 3C, after the feature region
94 is acquired, step 25 further acquires a location 940 in the
feature region 94 corresponding to a spot 930 in the region of
interest by geometric matching. Then, according to the location
940, the angular relation based on the coordinates of the feature
region 94 is obtained. Furthermore, normalization can be achieved
based on the angular relation (.theta.) and scale relation between
the feature region 94 and the region of interest 93.
[0040] Referring to FIG. 1, in step 206, the feature region in
other images is acquired repeatedly until the number of processed
car images reaches the number of plurality of images. Afterwards,
step 207 is performed to normalize the acquired feature regions.
The step of normalization is performed according to the angular
relation and the scale relation in step 205 so that the size of
each feature region is adjusted to be identical to the size of the
region of interest or that the sizes of the feature region and the
region of interest are adjusted to a specific scale. Since there
are a plurality of images acquired in step 200, the size of the
target (i.e., a car in the present embodiment) may vary, as shown
in FIG. 3A and FIG. 3B, due to different view angles and distances.
As a result, the feature regions 93 and 94 acquired in step 205 may
differ in size. Therefore, step 207 is performed to adjust the size
of each feature region to be identical. In the present embodiment,
the size is 130.times.130 pixels.
[0041] Then, step 208 performs an inversing operation on the pixels
in the plurality of feature regions and the pixels in the region of
interest respectively. The inversing operation is achieved so that
the bright turns dark and the dark turns bright. The sensitivity of
human eyes is saturated in a bright environment, which leads to
failure in identifying the detailed structure in the bright region.
Therefore, it is easier to identify the images after the inversing
operation is performed to turn the bright into dark. Afterwards,
step 209 is performed to form a recombined image by summing the
pixels in the plurality of feature regions and the pixels in the
region of interest. Since the feature region and the region of
interest have been normalized to have the same image size, the
greyscale values of the pixels can be summed. In step 210, an
inversing operation is performed on the recombined image. Then,
step 211 is performed to enhance the recombined image. The image is
enhanced to improve the contrast and the brightness. Referring to
FIG. 4A and FIG. 4B, FIG. 4A shows unidentified feature regions on
a plurality of images (3 images in the present embodiment) before
recombination while FIG. 4B is a schematic diagram of a recombined
image by the method shown in FIG. 1. In other words, the 3 images
in FIG. 4A are recombined to obtain a clear image.
[0042] Please refer to FIG. 5, which is a flowchart of a method for
image recombination of a plurality of images according to another
embodiment of the present invention. In the present embodiment,
step 300 to step 307 in the method 3 are similar to step 200 to
step 207 in FIG. 1, and descriptions thereof are thus not presented
herein. The method 3 in the present embodiment is different from
FIG. 1 in that the operation for obtaining the recombined image is
different. After normalization in step 307, step 308 is performed
to average the feature regions in the plurality of images to obtain
a recombined image. Each pixel in the feature region and region of
interest is summed and averaged to form the recombined image. Then,
in step 309, a histogram equalization process is performed on the
recombined image. Th histeogram equalization process is aimed at
enhancing the contrast of the recombined image. For example, FIG.
6A shows unidentified feature regions on a plurality of images (3
images in the present embodiment) before recombination, wherein the
features on each image are blur and unclear. After step 308 and
step 309 Are performed, a clear image is formed as shown in FIG.
6B. FIG. 7A and FIG. 7B show the greyscale value with respect to
the location before and after histogram equalization respectively.
It is observed that the contrast difference d is small (FIG. 7A)
before the histogram equalization process in step 309, while the
contrast difference D is large (FIG. 7B) after the histogram
equalization process in step 309. A larger contrast difference D
(FIG. 7B) is helpful for image identification. Referring to FIG. 5,
step 310 is performed to enhance the features in the recombined
image for image identification.
[0043] Please refer to FIG. 8, which is a flowchart of a method for
image identification according to the present invention. The
flowchart of the method for image identification can be achieved
using the recombined image formed in FIG. 1 or FIG. 5 to obtain
identification results with respect to the recombined image. In
other words, the recombined image formed in FIG. 1 or FIG. 5 is
used to obtain information according to the comparison with the
plurality of sample images. The method 4 starts with step 40 to
provide a database. The database provides a plurality of standard
sample images. Please refer to FIG. 9A, which is a schematic
diagram of a sample image. The size of the sample image 5 is
determined according to the user's demand, for example,
130.times.130 pixels, but not limited thereto. A standard image
region 50 is formed on the pixel in the sample image 5. The
standard image region 50 comprises a plurality of pixels 500 and
501 to form a character, a digit, a word or a pattern as
represented by the sample image. Referring to FIG. 9B, the present
embodiment is exemplified by a digit "1". In the sample image 5,
each pixel 500 and 501 is given a proper greyscale value to form a
standard image region 50, which draws the outline of the digit 1.
Then, in the standard image region 50, specific pixels 501 (pixels
with oblique lines) are given a specific weight value. The
greyscale value and the weight value are determined according to
the user's demand. That is, each weight value may be different or
identical. In the present embodiment, the weight value is positive.
In the standard image region 50, the greyscale value and the weight
value for each pixel 500 and 501 are combined as the first feature
value.
[0044] In the sample image, the non-standard image region 51 is
provided as shown in FIG. 9C. The non-standard image region 51
represents the content that the standard image region 50 is taken
for. For example, digit "1" is often taken for letter "I" or "L" or
even letter "E. Therefore, locations for pixels 510 possibly
mis-identified (pixels with dots) are given proper greyscale values
and weight values as the second feature values corresponding to
pixels 510. In the present embodiment, locations for the pixels 510
in the non-standard image region 51 are determined according to the
easily mis-identified character, digit or word in the standard
image region 50, which is not restricted. The greyscale values and
weight values are determined according to practical demand. In the
present embodiment, the weight values in the non-standard image
region 51 are negative.
[0045] As shown in FIG. 9D, which is a schematic diagram showing
another sample image 5a provided according to digit 0, the sample
image 5a also comprises a standard image region and a non-standard
image region. The pattern constructed by the pixels in the standard
image region draws the outline of a digit "0". Similarly, the
pattern constructed by the pixels in the non-standard image region
denotes a word that digit "0" is taken for. For example, digit "0"
is often taken for letter "Q" or digit "8". Steps 221 and 222 can
be performed using image processing software exemplified by, but
not limited to, MS Paint. The sample images, such as 0 to 9, A to Z
and a to z, are stored in the database. Then, in step 224, the
identification result is observed after a plurality times of
training.
[0046] Referring to FIG. 8, step 41 is performed to acquire a
feature image from the recombined image. For example, in FIG. 10A,
the region in the recombined image 95 (the image in FIG. 6B)
corresponding to each unidentified word is the feature image. In
step 41, the acquired feature image 96 is the first character in
the identification information. Then, step 42 performs a
calculation on a third feature value of each pixel in the feature
image and the first feature value or the second feature value
corresponding to each pixel in the plurality of sample images to
obtain a similarity index of the feature image corresponding to the
plurality of sample images respectively.
[0047] Please refer to FIG. 10B, which is a schematic diagram
showing a feature image 96. The feature image can be processed with
each of the sample images for further calculation to obtain a
corresponding similarity index C.sub.uv. The calculation is based
on normalized correlation matching, as described in equation (1).
Normalized correlation matching is aimed at calculating the
relation between the feature image and the sample image, wherein
the standard deviation of the greyscale value of each image is
regarded as a vector and is multiplied with the weight value so as
to determine the optimal location. The standard correlation value
is within the range between -1 and 1 with higher similarity as it
gets closer to 1. When C.sub.uv reaches its maximum, an optimal
location is achieved.
C uv = ( u i - u _ ) ( v i - v _ ) .times. w i [ ( u i - u _ ) 2 (
v i - v _ ) 2 ] 1 / 2 ( 1 ) ##EQU00001##
wherein u.sub.i is the greyscale value of each pixel in the sample
image, while v.sub.i is the greyscale value of each pixel in the
feature image, i.e., the third feature value. Moreover, is the
average greyscale value of all the pixels in the sample image,
while v is the average greyscale value of all the pixels in the
feature image. w.sub.i is the weight value of the pixels in the
standard image region and the non-standard image region in the
sample image. The weight value of pixels in the other region is
1.
[0048] Based on equation (1), a calculation is performed on each
pixel in FIG. 10B and each pixel in the sample image. For example,
FIG. 10B and the sample image in FIG. 9C (representing digit 1) and
the sample image in FIG. 9D (representing digit 0) are calculated
to obtain the similarity index C.sub.uv of the feature image in
FIG. 10B corresponding to FIG. 9C and FIG. 9D. Referring to FIG. 8,
after obtaining the similarity index, steps 43 and 44 are performed
to acquire the feature image from each character in the recombined
image 95. Step 42 is then repeated to perform identification. Step
45 collects a plurality of similarity indexes with respect to the
feature image compared with the plurality of sample images. In the
present step, the similarity indexes are sorted from the
identification result with highest possibility to the
identification result with lowest possibility. Finally, in step 46,
the plurality of similarity indexes are sorted and at least one of
comparison results is output.
[0049] Referring to FIG. 3B, since there are 7 characters in the
identification mark, the result shown in FIG. 11 can be obtained
after the flowchart of the method 4 for identification is
performed. In FIG. 11, four possible results are shown. Each result
represents one possible combination of characters on the license
plate. Each character in the first possible result has the highest
similarity, which is followed by the second, the third and the
fourth possible results. Taking the first possible result for
example, the characters on the license plate are possibly 6095-OA,
wherein the first character "6" has a similarity index of 72, the
second character "0" has a similarity index of 52, the third
character "9" has a similarity index of 67, the fourth character
"5" has a similarity index of 72, the fifth character is "-", sixth
character "O" has a similarity index of 63, and the seventh
character "A" has a similarity index of 76. Certainly, the user can
also determine other combinations of characters on the license
plate number according to the results in FIG. 11 and visual
estimation on the image to be identified.
[0050] In image identification, images of impossible characters or
digits can be excluded according to various combinations that form
the identification marks. For example, in one embodiment, the
identification mark can be formed as a combination of 4 leading
digits and 2 following letters (as shown in FIG. 3A) with a "-"
therebetween. In another identification mark, 2 leading letters and
4 follwing digits are combined, with a "-" therebetween. In the
present embodiment, there are two kinds of combinations to
exemplify the license plates. Therefore, images of impossible
characters or digits can be excluded according to the relative
locations of the feature images in the identification mark so as to
increase identification efficiency.
[0051] Please refer to FIG. 12, which is a schematic diagram of a
system for image acquiring and identification according to the
present invention. The system 6 is capable of implementing the
flowchart in FIG. 1, FIG. 5 or FIG. 8 for image identification and
identification result output. The system 6 comprises a database 60,
an image processing unit 61, an identification and output unit 62,
a plurality of image acquiring units 63 and an image input unit 64.
The database 60 is capable of providing a plurality of sample
images. The plurality of image acquiring units 63 are electrically
connected to the image processing unit 61. Each image acquiring
unit 63 is capable of acquiring an image of an object and transmits
the image to the image processing unit 61 for identification. In
the present embodiment, each of the image acquiring units 63 is
capable of acquiring dynamic or static images of the object. The
image provides an identification region for carrier identification.
The identification region comprises identification information. The
image acquiring units may be CCD or CMOS image acquiring units, but
not limited thereto. The object may be a carrier with an
identification mark thereon, for example, the license plate number
of a car. Moreover, the object may also be a word, a character, a
digit or combinations thereof.
[0052] The image input unit 64 is capable of receiving and
transmitting the plurality of images acquired by the image
acquiring unit 63 to the image processing unit 61. The image
processing unit 61 comprises a feature acquiring unit 610, a
recombination unit 611, an enhancing unit 612 and an identification
and comparison unit 613. The feature acquiring unit 610 is capable
of acquiring features in the region of interest in a standard image
and acquiring a feature region corresponding to the region of
interest according to acquired features in other regions. The
standard image is formed by choosing one image from the plurality
of images. The recombination unit 611 performs an image
recombination process according to the plurality of feature regions
and the region of interest to form a recombined image. The
recombined image is formed as disclosed in FIG. 1 or FIG. 5. The
enhancing unit 612 is capable of improving the recombined image to
enhance the contrast, brightness or the edge features of the
recombined image.
[0053] The identification and comparison unit 4111 performs step 23
in FIG. 1 to compare the feature image with the sample image to
obtain the plurality of similarity indexes corresponding thereto,
and sorts the plurality of similarity indexes to output at least
one of comparison results. The identification and output unit 42 is
electrically connected to the processing unit 41 to output the
comparison result identified by the processing unit 41. The output
from the identification and output unit 42 is as shown in FIG. 8A,
which is capable of allowing the user to know the identification
results displayed on a display.
[0054] The identification and comparison unit 613 is electrically
connected to the enhancing unit 612 to identify the recombined
image. The identification and comparison unit 613 performs a
calculation on each pixel in the feature image and each pixel in
the plurality of sample images to obtain a similarity index of the
feature image corresponding to the plurality of sample images
respectively according to the flowchart in FIG. 8, and further
collects a plurality of similarity indexes with respect to the
feature image compared with the plurality of sample images. The
output unit 62 is electrically connected to the processing unit 61
to output at least one of comparison results from the processing
unit 61.
[0055] Accordingly, the present invention discloses a method for
image processing and identification and a system for image
acquiring and identification with enhanced efficiency and
precision. Therefore, the present invention is useful, novel and
non-obvious.
[0056] Although this invention has been disclosed and illustrated
with reference to particular embodiments, the principles involved
are susceptible for use in numerous other embodiments that will be
apparent to persons skilled in the art. This invention is,
therefore, to be limited only as indicated by the scope of the
appended claims.
* * * * *