U.S. patent application number 10/673281 was filed with the patent office on 2004-04-01 for method of searching or browsing multimedia data and data structure.
This patent application is currently assigned to LG ELECTRONICS, INC.. Invention is credited to Kim, Hyeon Jun, Lee, Jin Soo.
Application Number | 20040064526 10/673281 |
Document ID | / |
Family ID | 26634655 |
Filed Date | 2004-04-01 |
United States Patent
Application |
20040064526 |
Kind Code |
A1 |
Lee, Jin Soo ; et
al. |
April 1, 2004 |
Method of searching or browsing multimedia data and data
structure
Abstract
A method of searching or browsing multimedia data which can
effectively search or browse an image is disclosed. In the present
method, the searching or browsing is performed in consideration of
at least one of a reliability information on a weight of a feature
of an input multimedia data and an authority code. According to the
method, the system selects reference multimedia data, and searches
the multimedia data using a weight of a feature and/or a feature
element of the input multimedia data. A user gives feedback on the
relevance of the searched or browsed multimedia data, and the
system calculates a new weight using the relevance information.
Thus, the system updates and maintains new weights of features
and/or feature elements.
Inventors: |
Lee, Jin Soo; (Seoul,
KR) ; Kim, Hyeon Jun; (Pundang-gu, KR) |
Correspondence
Address: |
FLESHNER & KIM, LLP
P.O. Box 221200
Chantilly
VA
20153-1200
US
|
Assignee: |
LG ELECTRONICS, INC.
|
Family ID: |
26634655 |
Appl. No.: |
10/673281 |
Filed: |
September 30, 2003 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
10673281 |
Sep 30, 2003 |
|
|
|
09493038 |
Jan 28, 2000 |
|
|
|
6643643 |
|
|
|
|
Current U.S.
Class: |
709/217 ;
707/E17.02; 707/E17.023; 707/E17.031 |
Current CPC
Class: |
Y10S 707/99936 20130101;
Y10S 707/99939 20130101; G06F 16/54 20190101; G06F 21/00 20130101;
G06F 16/583 20190101; G06F 16/51 20190101; G06F 16/5838 20190101;
Y10S 707/99935 20130101; Y10S 707/99932 20130101 |
Class at
Publication: |
709/217 |
International
Class: |
G06F 015/16 |
Foreign Application Data
Date |
Code |
Application Number |
Jan 29, 1999 |
KR |
2979/1999 |
Aug 27, 1999 |
KR |
35798/1999 |
Claims
What is claimed is:
1. A method of searching or browsing multimedia data comprising:
(a) receiving a reference multimedia data with data structure
including features of said reference multimedia data and weight
information of said features; (b) searching for said reference
multimedia data using the features and the weight information; (c)
receiving user feedback on a relevance of resultant multimedia data
found in (b); (d) measuring a similarity of the reference
multimedia data to the resultant multimedia data and calculating a
new weight information of said features using the measured value;
and (e) updating the weight information of said features in said
data structure of the reference multimedia data using the new
weight information.
2. A method of claim 1, wherein in (c), increasing weights of
features which would increase a similarity between the reference
multimedia data and the resultant multimedia data if the user
feedback is a positive relevance information
3. A method of claim 2, wherein in (c), increasing weights of
features which would increase a dissimilarity between the reference
multimedia data and the resultant multimedia data if the user
feedback is a negative relevance information.
4. A method of claim 1, wherein the data structure of the reference
multimedia data further comprises reliability information
indicating a reliability of the weight information.
5. A method of claim 4, wherein a reliability of a weight assigned
to a feature is proportional to the amount of training by user
feedback.
6. A method of claim 4, wherein the data structure of the reference
multimedia data further comprises authority information which
limits an update of the weight information by a user feedback.
7. A method of claim 6, wherein the authority information includes
a plurality of authority levels, wherein each authority level has
degree values affecting the degree of weight information update in
(e).
8. A method of claim 7, wherein a higher reliability is given to
user feedback by a user with a high authority level.
9. A method of claim 6, wherein in (e), updating the weight
information of said features in said data structure of the
reference multimedia data depending upon the reliability
information and the authority information.
10. A method of claim 1, wherein the data structure of the
reference multimedia data further comprises authority information
which limits an update of the weight information by a user
feedback.
11. A method of claim 10, wherein the authority information
includes a plurality of authority levels, wherein each authority
level has degree values affecting the degree of weight information
update in (e).
12. A method of claim 11, wherein receiving a password from a user
to determine an authority level of the user.
13. A method of searching or browsing multimedia data comprising:
searching for a reference multimedia data using variable
information representing an importance of a feature of the
reference multimedia data; receiving user feedback on relevance of
the searched multimedia data; calculating a new variable
information using the relevance information as training information
or using user pattern information stored in the system; and
updating the variable information using the calculated new variable
information and/or reliability information, and maintaining the
updated variable information.
14. A method of claim 13, wherein the variable information is one
or a combination of: a weight of features used for searching for
the reference multimedia data; a weight of a frame or segment used
for searching or browsing a specific video; a weight of
user-dependent information representing a user preference or user
habit and a portion of data used for searching or browsing a main
region or main object of the multimedia data; and information for
multimedia grouping of a similarity list and a cluster model.
15. A method of claim 13, wherein if at least two identical
multimedia data having different variable information for a search
or browsing are provided, selecting a multimedia data which has a
relatively high reliability with respect to the variable
information to calculate the new variable information.
16. A method of claim 13, wherein if at least two identical
multimedia data having different variable information for a search
or browsing are provided, combining the provided multimedia data in
proportion to the reliability of the variable information value to
calculate the new variable information.
17. A method of searching or browsing multimedia data comprising:
searching for a reference multimedia data using variable
information representing an importance of a feature of the
reference multimedia data; receiving user feedback on relevance of
the searched multimedia data; and updating the variable information
using the relevance information as training information or using a
prestored user pattern information, and/or reliability information,
and maintaining the updated variable information.
18. A data structure for a multimedia data searching or browsing
system comprising: a multimedia data; and a variable information
representing features of the multimedia data.
19. A data structure of claim 18, further comprising a reliability
information representing a reliability of the variable
information.
20. A data structure of claim 19, wherein the reliability
information includes information on the number of variable
information updates by a user.
21. A data structure of claim 19, further comprising an authority
code.
22. A data structure of claim 21, wherein the reliability
information is variable and includes a number of authority levels,
a degree of variable information update for each authority level,
and a number of variable information updates by a user of each
authority level.
23. A data structure of claim 21, wherein the reliability
information is fixed and includes a number of variable information
updates by a user of fixed authority levels.
24. A data structure of claim 18, further comprising an authority
code.
25. A method of searching or browsing a multimedia data having a
data structure as in claim 18.
26. A method of searching or browsing a multimedia data having a
data structure as in claim 19.
27. A method of searching or browsing a multimedia data having a
data structure as in claim 21.
28. A method of searching or browsing a multimedia data having a
data structure as in claim 24.
Description
BACKGROUND OF THE INVENTION
[0001] 1. Field of the Invention
[0002] The present invention relates to a method of searching or
browsing data, and more particularly to a method of searching or
browsing multimedia data such as an image or video.
[0003] 2. Background of the Related Art
[0004] Recently, technology for digital image signal processing has
been greatly developed and has been applied in various fields. For
example, the digital image signal processing technology may be used
in a search system for automatically editing only the face of a
specific character in a moving picture file of a movie or drama, in
a security system for permitting access to only persons registered
in the system, or in a search system for searching a particular
data from an image or video detected by a detection system. The
performance of such systems basically depend on the accuracy and
speed of detecting or searching the desired object. Accordingly,
various image searching methods have been proposed in the related
art.
[0005] An image search system which detects a degree of similarity
with an image to be searched utilizing features such as color,
texture or shape is disclosed in U.S. Pat. No. 5,579,471 entitled
"An image query system and method." Depending upon the image to be
searched, the importance of a feature may vary and within one
particular feature such as the color, the importance of a feature
element such as the red or green color may also vary. However, the
above searching system does not take into consideration the
different importance of features or feature elements for each image
to be searched.
[0006] In another searching method entitled "Virage image search
engine" (www.virage.com), a user directly inputs the level of
importance for features such as the color, texture and shape by
assigning weight values. Although an image may be searched
according to an importance of a feature using this method, it may
be difficult for a user to determine the importance of
features.
[0007] Therefore, Yong Rui in "Relevance feedback techniques in
interactive" SPIE Vol.3312, discloses a method in which images
similar to a reference image are found and the importance of
features or weight for features are automatically obtained by
calculating the similarities among the found images. However, the
weight importance information is not maintained after a search for
a specific image is finished and must be calculated for each image
search, even for a same image.
[0008] In the image search and browsing system or the video (moving
image) search and browsing system of the related art, information
which describes a particular feature of an image or video data is
utilized to perform a more effective search or browsing of the
multimedia data. For example, in the image query system, an image
may be divided into a plurality of regions and a representative
color of each region may be utilized as a feature information of
the image, or a whole color histogram of the image may be utilized
as a feature information. Thereafter, two images are compared to
calculate a similarity based upon the feature information and a
determination is made whether the two images are similar.
[0009] Also, a user may wish to search and view a section of a
video in which a specific character appears or a specific event
occurs. At this time, a key frame representing a specific event may
be utilized to perform a search and browsing of the video.
Moreover, in recent searching and browsing methods, a key frame and
a key segment have been defined utilizing an upper level feature
information and a lower level feature information. This type of
multimedia data search and browsing method may use weight
information which indicates a degree of importance or may directly
or indirectly use group information which was obtained in advance
by grouping similar multimedia data. Such weight information or
group information may be predefined in advance by an expert or may
be a feature information variably modified by user feedback.
[0010] Furthermore, when browsing the multimedia data such as an
image or video, the whole or a portion of the original multimedia
data can be displayed depending upon a user request or upon a
resource condition of the hardware. When portions of the original
multimedia data are displayed, weight values are assigned to each
portion such that an important portion among the whole data may be
preferentially displayed. This weight value may be a varied by a
user feedback or defined in advance by an expert.
[0011] In U.S. Pat. Nos. 5,020,019 and 5,297,042, information which
describes a user preference is utilized to provide a better service
for each user. For example, for a movie or video data, the
information may include user preference such as action movie, drama
or sports. In another example, users may have a different
preference and meaning for keywords used most frequently in a
keyword search. Therefore, each user can perform a more effective
search if weights of keywords used often by a user is assigned
differently according to the user preference. Such user preference
may be obtained using search conditions and history of previous
searches by the user. The information which depends on a user
preference may also be varied and updated.
[0012] Although the variable feature information may be sufficient
for some searching and browsing of multimedia data, when the
feature information which includes the weight values is predefined
by an expert, the reliability may vary depending upon the
trustworthiness of an expert and/or a number of expert opinions
used in defining the variable information. In other words, even if
the feature information describing a feature of the multimedia data
is defined in advance by an expert group, the reliability of the
feature information defined for a same image may be inconsistent
depending upon the skill of the expert. Similarly, when the feature
information is trained by a user feedback, the reliability may vary
depending upon the number and accuracy of feedbacks. For example,
if two images are incorrectly judged to be similar by a user after
a search, such information would be fed back and adversely affect
the training of the feature information.
[0013] In addition, a user preference input by a new user may be
different from the feature information recommended by a server. In
such case, the system should assign appropriate weights to the user
preference information depending upon the reliability of the user.
Finally, as different users gives different feedback to the system,
if the reliability of the existing feature information is
relatively high, individual feedback of a new user would have a
small effect on the update of the feature information. However, if
the reliability is low, a feedback of each new user would have a
large effect on the feature information value.
[0014] Since the searching and browsing methods of the related art
do not take into consideration the factors described above, either
the reliability of the searched result would be inconsistent or
reliable feedback must be input at all times.
SUMMARY OF THE INVENTION
[0015] Accordingly, an object of the present invention is to solve
at least the problems and disadvantages of the related art.
[0016] An object of the present invention is to provide an accurate
and effective method of searching or browsing multimedia data.
[0017] Another object of the present invention is to provide a
method of searching or browsing in consideration of a information
reliability and an authority code to perform an update of the
feature information.
[0018] A further object of the present invention is to provide a
data structure for use in searching or browsing multimedia
data.
[0019] Additional advantages, objects, and features of the
invention will be set forth in part in the description which
follows and in part will become apparent to those having ordinary
skill in the art upon examination of the following or may be
learned from practice of the invention. The objects and advantages
of the invention may be realized and attained as particularly
pointed out in the appended claims.
[0020] To achieve the objects and in accordance with the purposes
of the invention, as embodied and broadly described herein, the
method of searching or browsing multimedia data comprises selecting
or inputting a reference multimedia data; searching for the
reference multimedia data using weights assigned to features and/or
feature elements of the multimedia data; inputting user feedback on
the relevance of the searched or browsed multimedia data; measuring
a similarity of the reference multimedia data to the resultant
images of the search and calculating a new weight using a measured
value; and updating previous weights and maintaining the updated
weights.
[0021] In another aspect of the present invention, a method of
searching or browsing multimedia data comprises searching a
reference multimedia data using variable information representing
importance of features of multimedia data, said variable
information being included within the multimedia data structure,
and using a reliability information representing a reliability of
the variable information; inputting a user feedback on the
relevance of the searched or browsed multimedia data; calculating a
new variable information using training information fed back by a
user or user pattern information stored in the system; and updating
the previous variable information using the calculated variable
information and/or reliability information, and maintaining the
updated information.
[0022] In still another aspect of the present invention, a method
of searching or browsing multimedia data comprises selecting of
inputting a reference multimedia data; searching for the reference
multimedia data using variable information representing importance
of features of the multimedia data wherein the multimedia data
structure includes the variable information, a reliability
information of the variable information and an authority code for
limiting an authority for a user; inputting a user feedback on a
relevance of the searched or browsed multimedia data; measuring a
similarity of the reference multimedia data to the resultant images
of the search and calculating a new variable information using the
measured value; and updating the previous variable information
using a selective combination of the reliability information and
authority code, and maintaining the updated information.
[0023] In still a further aspect of the present invention, a data
structure for a multimedia data searching or browsing system
comprises a multimedia data; feature information of the multimedia
data; and weight information representing a weight of a feature of
the multimedia data.
BRIEF DESCRIPTION OF THE DRAWINGS
[0024] The invention will be described in detail with reference to
the following drawings in which like reference numerals refer to
like elements wherein:
[0025] FIG. 1 is a multimedia data structure in a multimedia data
searching or browsing system according to the present
invention;
[0026] FIG. 2 is a another data structure for use in a multimedia
data searching or browsing system according to the present
invention;
[0027] FIG. 3 is a flowchart of a multimedia data searching or
browsing process according to a first embodiment of the present
invention;
[0028] FIG. 4 is a data structure of variable reliability
information in a multimedia data searching or browsing system
according to a third embodiment of the present invention;
[0029] FIG. 5 is an example of a variable weight information in
multimedia data according to the data structure of FIG. 4;
[0030] FIG. 6 is a data structure of a fixed reliability
information in a multimedia data searching or browsing system
according to a third embodiment of the present invention; and
[0031] FIG. 7 is an example of the fixed weight information in
multimedia data according to the data structure of FIG. 5.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0032] Reference will now be made in detail to the preferred
embodiment of the present invention, examples of which are
illustrated in the accompanying drawings.
[0033] FIG. 1 is a multimedia data structure which may be used in a
search or browsing system according to the present invention,
including an image description for relevance feedback 101. The
image description for relevance feedback 101 is information which
indicates an importance of features or feature elements
(hereinafter generally referred to as "features") trained by user
feedback and includes an image description 102, an authority code
104 and a reliability code 107.
[0034] Particularly, the image description 102 includes image
features 103a for use in the actual search and weights 103b
indicating the importance of image features. The reliability code
107 indicates the reliability of the weights. In the preferred
embodiment, the reliability code is represented using the number of
weight updates by a user of level i 108. The authority code 104
indicates an authority level 106 and the number of the authority
levels 105.
[0035] The present invention utilizes the authority code 104 to
determine whether the weight should be updated and if so, the
degree of the update. Moreover, the reliability code 107 and the
authority code 104 may be selectively or both included in the data
structure depending upon the needs of the system. For example, the
data structure may include only the reliability code 107 and the
image description 102. In such case, the system may provide the
authority code 104. Also, if only one authority level is required
in the system, the authority code (104) may be omitted.
[0036] The weight 103b of the image description. 102 may be updated
upon a user feedback and is the weight values assigned to features
of the image. Furthermore, an authority level i 106 of the
authority code 104 determines whether to update a weight assigned
to an image feature and/or the degree of the update. The number of
different levels may be determined based upon the needs of the
system and is indicated in the number of authority levels 105. For
example, assuming there are five levels, a weight may be updated
upon feedback by users of the first and second authority levels,
and the degree of update for the first level user may be 0.95 while
the weight for the second level user may be 0.85.
[0037] FIG. 2 is a view illustrating a data structure for use in a
multimedia data searching or browsing system according to the
present invention. Referring to FIG. 2, the data structure includes
a multimedia data 201, a feature information 202 of the multimedia
data, a weight information (or variable information) 203
representing the importance of features of the multimedia data, a
reliability information 206 and 208 representing the reliability of
the weight information, and an authority code representing an
authority level for a user.
[0038] As described with reference to FIG. 1, the authority levels
are used to determine whether to update a weight information and
the degree of such update. Thus, in the data structure, the
authority code includes the number of authority levels 204 and
authority level passwords 205 and 207 for the authority levels.
Different passwords may be given to users based upon the authority
level to limit or permit appropriate image access, i.e. access to
modify the data, based on the authority code. Accordingly, an image
access would not be permitted to the users having an authority
levels lower than a specified level.
[0039] In the preferred embodiment, an image producer (or creator)
divides the user authority into n levels and gives appropriate
passwords to users based upon the authority level. Thereafter, a
user can search or browse images and vary the image description of
an image according to a given authority level.
[0040] Moreover, because the image description is updated by users
of different authority levels, the number of feedbacks by users of
each authority level is recorded to calculate the reliability
information. Specifically, the number (Nk) of weight updates for an
image description by user feedbacks of the kth level is multiplied
with a weight coefficient (ak) assigned to the kth level. The
product values (ak*Nk) for each respective authority level (1-n)
are added and the sum is multiplied by a predetermined reliability
coefficient (a) to obtain the reliability of the image description.
If two input data have the same multimedia data, but different
image description trained by different histories, the user can
determine which image to use based upon the reliability information
in the respective image description. Here, if there is only one
authority level, the image would have a reliability value without
the authority information.
[0041] In the first embodiment of the present invention, the
multimedia data searching or browsing method is explained when
variable information describing features of an image and/or an
authority code is included in the input multimedia data. FIG. 3 is
a flowchart of a searching or browsing method according to a first
embodiment of the present invention. For purposes of explanation,
the multimedia data will be assumed to be an image.
[0042] Referring to FIG. 3, a reference image is selected or input
to the searching or browsing system by a user (step S301). The
system then searches for similar images using image features and
weights assigned to image features (step S302). The system
determines an initial values of the weight if the reference image
is selected or input for the first time. The weight values may be
assigned for image features such as color, texture or shape and/or
may be assigned for image feature elements such as the Ith color
element.
[0043] Upon viewing the search or browsing result, the user may
give feedback by inputting relevance information, i.e. whether the
search or browsing result is relevant. Particularly, the user may
give positive feedback if the result is good or give negative
feedback if the result poor. For example, images determined by a
user to be similar to the reference image would be fed back as
positive feedbacks and images determined by a user to be different
from the reference image would be fed back as negative
feedbacks.
[0044] Thus, the system judges whether the user has input relevance
information (step S303) and if there is no user feedback, the
system returns to step S301. However, if there is user feedback,
the system determines whether the user has sufficient authority
level to update the weight of features (step S304). Here, the
system may determine whether the authority level of the user by
requiring a user to enter a password. A detailed explanation of the
authority levels will be explained below with reference to the
third embodiment. Generally, step S304 limits the update of the
weight including the degree of the update depending upon the
authority level. However, such step may be omitted, in which case,
the weight value would be updated for each and every user
feedback.
[0045] Referring back to FIG. 3, if the authority level of the user
allows an update, the system calculates the new weights by
measuring the similarity of the reference image and the fed back
images (steps S305 and S306) among the searched images. Namely, the
system increases the weights of features which would increase the
similarity between the reference image and the fed back images if
the relevance information is positive information. On the contrary,
if the relevance information is negative information, the system
increases the weights of features which would increase the
dissimilarity between the reference image and the fed back
images.
[0046] Thereafter, the system updates the previous weight
information in the image description using the new weights and
maintains the updated weights for the next search or browsing (step
S307) Particularly, a weight value is updated by taking into
consideration the weight of the kth feature prior to the update,
the new weight value of the kth feature obtained in the current
image search and an average of new weight values of all features.
Here, the degree of effect on the updated weight value due to the
previous weight, i.e. weight of the kth feature prior to the
update, is inversely proportional to the reliability of the
previous weight.
[0047] In the preferred embodiment, the weight value can be updated
using Equation 1, where Wk(t) is the updated weight of the kth
feature, Wk(t-1) is the weight of the kth feature prior to an
update, Wnew_k is the calculated weight of the kth feature obtained
by a current image search, M(Wnew) is the average of the calculated
weights for all features obtained by the current image search, and
a is a weight coefficient which determines the degree of update
according to the user authority level.
Wk(t)=Wk(t-1)+.DELTA.Wk, (.DELTA.Wk=a(Wnew.sub.--k-M(Wnew)) [1]
[0048] Alternatively, the weight may be updated using a similar
image list or a dissimilar image list obtained during the search
process. For example, the system would measure the similarity among
the images in the similar list and increase the weight of features
contributing to the judgement of the similarity. On the contrary,
the system measures the dissimilarity among the images in the
dissimilar image list, and increase the weight of features
contributing to the judgement of the dissimilarity.
[0049] According to the first embodiment of the present invention,
the weight information of features used in a search is included
within the multimedia data structure input to the system. Thus, the
system develops a proper weight by self-training during each
searching or browsing of an image. As a result, the system can
perform a more effective search or browsing using the weight
information of features irrespective of a place or application
program, image searching engine. Also, the present embodiment may
be adapted and used for searching or browsing an image that the
user desires by performing an effective response to subjective and
objective queries by the user whenever weights are used for a
search or browsing. Accordingly, the first embodiment of the
present invention can be suitably adopted to be used as an image
format that can use the "relevance feedback" concept for
controlling the weight according to the user feedback.
[0050] In a second embodiment of the present invention, the
reference data selected or input for searching or browsing includes
variable information representing the importance of features of the
multimedia data and reliability information representing the
reliability of the variable information. In the second embodiment,
the multimedia data structure which includes the variable
information and the reliability information is input to the system,
and the system searches for the reference data. For purposes of
explanation, the multimedia data will be assumed to be an
image.
[0051] As in the first embodiment, the user may give feedback on
the relevance of the resultant images of the search or browsing.
The system then calculates a new variable information using either
the relevance information input by the user or the user pattern
information stored in the system. For example, the user pattern
information may be a history of a user or the history of uses.
Thereafter, the system updates the variable information using the
new variable information and/or the reliability information, and
maintains the updated variable information. Here, the variable
information is updated in an analogous manner as the update of
weights explained with reference to FIG. 3.
[0052] The variable information may be a weight assigned to
features used for searching such as the color, texture, or keyword
of the multimedia data; a weight assigned to a frame or segment
used in searching or browsing a specific video; a weight assigned
to user-dependent information representing a user preference or
habit, and representing a portion of data used for searching or
browsing a main region or main object of the multimedia data; a
information for a multimedia grouping of either a similarity list
or a cluster model; or a combination of two or more weights as
described above.
[0053] Generally, the reliability information directly or
indirectly represents the reliability of the variable information
and is proportional to the amount of user training of the variable
information according to the input relevance information. Also, the
reliability information can be determined by the performance of the
system. A detailed explanation of the reliability information will
be explained below with reference to the third embodiment.
[0054] Accordingly, if two data has identical multimedia data but
different variable information values, the system may select the
variable information with the higher reliability or combine the
variable information in proportion to the reliability of the
variable information when calculating a new variable information.
The calculation of the variable information may be used
independently irrespective of the second embodiment of the present
invention.
[0055] For example, assume that an image is created and is copied
into a plurality of images by other users, and further assume that
the image description of each copied image is developed differently
while passing through different application programs. Particularly,
the image description developed by different application programs
will have different variable information and reliability
information of the variable information. If the system then
requires a use of the image in a search, a determination as to
which copy of the image to use must be made. Because the images
with different image description have the reliability information
of the variable information, i.e. weight, the system will be able
to easily select and use the image having the highest reliability
for the search. Also, the system may obtain a new weight by
combining the weights of the different image descriptions
corresponding to the plurality of image copies in proportion to
their reliabilities.
[0056] According to the second embodiment of the present invention
as described above, since the reliability information of the
variable information as well as the variable information is
included in the image input to the system, the weights may be
updated whenever the image is searched. Thus, the variable
information is trained, allowing a more accurate search element. As
a result, the system may intellectually develop to an optimal
state.
[0057] In a third embodiment of the present invention, the
multimedia data structure input to the searching or browsing system
includes the variable information representing the importance of
features, the reliability information representing the reliability
of the variable information, and the authority code for limiting an
authority for a user. Here, the authority code may be included in
the multimedia data or may be provided from the system when
justification of the multimedia data search or browsing is fed
back.
[0058] In the third embodiment of the present invention, the
multimedia data search and browsing system selects or inputs a
reference multimedia data according to a control signal from the
user. Thereafter, the system searches the multimedia data using the
variable information, the reliability information of the variable
information, and the authority code. The authority code is used to
limit an improper training of the variable information by updates
in response to all user feedbacks regardless of the reliability of
such feedbacks.
[0059] For example, an inexperienced user may consider two
different images as similar causing the weight assigned to features
of the image used in the search to be updated in an improper
direction. Thus, the training result of the corresponding image
would be lower in reliability. However, in an alternative method,
the input image may not include the authority code, instead the
system may limit or discriminate the authority of weight updates or
the degree of the weight update by providing the authority code.
For example, user Log-In IDs may be utilized to determine the
authority level.
[0060] Moreover, as in the first and second embodiments, the user
gives feedback on the relevance of the searched or browsed
multimedia data. Thereafter, the system measures the similarity of
the reference image to the fed back images among resultant images
of the search, and calculates new variable information using the
measured value. The system then updates the previous variable
information included in the reference image using the calculated
variable information, reliability information and/or authority
code, and maintains the updated variable information. Thus, if
there are two images having different variable information values,
the system selects one which has a higher reliability with respect
to the variable information value.
[0061] Particularly, the new weights may be calculated and updated
as in the first embodiment by Equation 1, or by a similarity list
or dissimilarity list obtained during the search process.
Furthermore, the weight may be updated based upon the reliability
of the weight. The weight can be updated by Equation 2 where (Rc)
is a built-up reliability of the currently trained image, (Wc) is a
weight predetermined for the reliability, (Rt) is the degree of
updating the weight corresponding to the currently fed back user
authority level, and (Wt) is the weight for the user authority
level.
Updated weight=(RcWc+RtWt)/(Rc+Rt) [2]
[0062] Here, the value of Rc is the number of previous weight
updates by a user of respective authority levels.
[0063] FIG. 4 shows a data structure of a variable reliability
information in the multimedia data searching or browsing system
according to the third embodiment of the present invention.
Referring to FIG. 4, the variable reliability descriptor 401 is
organized by the number of levels 402, the degree of effect of
level I 403, and the number of updates by a user of level I. In
contrast, FIG. 6 shows a data structure of a fixed reliability
information in the multimedia data searching or browsing system
according to the third embodiment of the present invention.
Referring to FIG. 6, the fixed reliability descriptor 601 includes
the number of updates 602 performed by the user of level I. The
reliability information can be obtained as follows.
[0064] First, a higher reliability is given to an image with
variable information trained frequently by users having high
authority level. In other words, the reliability is given in
proportion to the number of the users who give feedback and the
authority level. Accordingly, the system gives a relatively high
reliability to images which has a relatively large amount of
training by relevance information fed back by users having high
authority level.
[0065] Second, the system gives a higher reliability to an image
trained mostly by users having authority level in the range from a
predetermined level to the upper level. In other words, the system
gives a relatively high reliability to images having a relatively
large amount of training by relevance information fed back by the
user having an authority level over a predetermined level.
[0066] Third, the reliability can be determined according to the
performance of the system.
[0067] Fourth, the system may use the reliability information as
shown in FIG. 4 by variably setting and using a number of user
levels with corresponding degree of update, or may use a fixed
reliability information as shown in FIG. 6. In any case, a
relatively higher reliability would be given as the number of
updates by user feedbacks and the authority level of the user
giving feedback becomes higher.
[0068] FIG. 5 shows an example of a variable reliability
information of FIG. 4. In the example, users are classified into
five authority levels. The first level has an update degree of 1.0
and has a record of eight updates by users of the first level. The
second level has an update degree of 0.7 and has a record of
fourteen updates by users of the second level. The third level has
an update degree of 0.5 and has a record of thirty updates by users
of the third level. The fourth level has an update degree of 0.3
and has a record of twenty-three updates by users of the fourth
level. The fifth level has an update degree of 0.1 and has a record
of four updates by users of the fifth level. Thus, the number of
authority levels and the degree of update influenced by users of
the respective authority level are variably determined.
[0069] In contrast, FIG. 7 is an example of a fixed reliability
information of FIG. 6. Referring to FIG. 7, the number of user
authority level is fixed to 10 levels and a corresponding
weight-updating degree for each level is allocated in advance from
the range of 0.1 to 1. Particularly, the first level has an update
degree of 1.0 and has a record of eight updates, the fourth level
has an update degree of 0.7 and has a record of fourteen updates,
the sixth level has an update degree of 0.5 and has a record of
thirty updates, the eighth level has an update degree of 0.3 and
has a record of twenty-three updates, and the tenth level has an
update degree of 0.1 and has a record of four updates.
[0070] The system checks the user authority level as determined
above, and effectively controls the access of the image, i.e.
permission and/or prohibition of weight update or degree of update.
Also, in the system shown in FIG. 7, all or a portion of the 10
levels may be used as necessary. For example, the system may use
five levels as in FIG. 5 by not assigning the remaining five levels
to users. In such case, the first level of FIG. 7 corresponding to
the first level of FIG. 5 would be the first level, the fourth
level of FIG. 7 corresponding to the second level of FIG. 5 would
be the second level, the sixth level of FIG. 7 corresponding to the
third level of FIG. 5 would be the third level, the eight level of
FIG. 7 corresponding to the fourth level of FIG. 5 would be the
fourth level, and the tenth level of FIG. 7 corresponding to the
fifth level of FIG. 5 would be the fifth level. Accordingly, 5
authority levels among 10 fixed authority levels are used and the
system can determine whether to permit an update based upon the
respective authority levels.
[0071] According to the embodiments of the present invention as
described above, the following effects can be provided.
[0072] First, since the weights (or variable information) of image
features are included in the input image and the corresponding
image is trained to provide a better search result when the image
search is repeated, an effective and accurate search or browsing
can be achieved when performing the search with respect to equal
images.
[0073] Second, when the equal images are updated after being
trained in different environments, the reliability information
representing the reliability of the weight or variable information
of the image is included in the respective image so as to identify
which image has been more effectively trained. Thus, the user can
search or browse the current image effectively and accurately.
[0074] Third, the permission of update and the degree of update are
limited according to user authority levels by including an
authority code in the input image or by providing the authority
code by the system. Accordingly, the corresponding image is
prevented from being improperly trained due to thoughtless or
abnormal updates by a user.
[0075] Fourth, since the reliability information and the authority
code is combined with the weight (or variable information) of the
image features selectively or together according to the
characteristic of the system, the multimedia data search or
browsing can be properly performed in a direction where the
intelligent training and self-development are effected.
[0076] The foregoing embodiments are merely exemplary and are not
to be construed as limiting the present invention. The present
teachings can be readily applied to other types of apparatuses. The
description of the present invention is intended to be
illustrative, and not to limit the scope of the claims. Many
alternatives, modifications, and variations will be apparent to
those skilled in the art.
* * * * *