Relevance maximizing, iteration minimizing, relevance-feedback, content-based image retrieval (CBIR)

• US 6,748,398 B2
• Filed: 03/30/2001
• Issued: 06/08/2004
• Est. Priority Date: 03/30/2001
• Status: Expired due to Term

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Su, Zhong, Zhang, Hong-Jiang, Zhu, Xingquan
• Primary Examiner(s): Metjahic, Safet
• Assistant Examiner(s): AL HASHEMI, SANA A
• Application Number:
  US09/823,534
• Publication Number:
  US 20020174120A1
• Time in Patent Office: 1,166 Days
• Field of Search: 707/5, 707/2, 707/3, 707/10, 707/104.1, 707/6, 709/203, 382/104, 382/165
• US Class Current: 1/1
• CPC Class Codes: G06F 16/5838 using colour
• CPC Class Codes: G06F 18/2178 based on feedback of a supervisor
• CPC Class Codes: G06F 18/22 Matching criteria, e.g. proximity measures
• CPC Class Codes: G06F 18/24155 Bayesian classification
• CPC Class Codes: G06F 18/40 Software arrangements specially adapted for pattern recognition, e.g. user interfaces or toolboxes therefor
• CPC Class Codes: G06V 10/761 Proximity, similarity or dissimilarity measures
• CPC Class Codes: G06V 10/7784 based on feedback from supervisors
• CPC Class Codes: G06V 10/945 User interactive design; Environments; Toolboxes
• CPC Class Codes: Y10S 707/99931 Database or file accessing
• CPC Class Codes: Y10S 707/99932 Access augmentation or optimizing
• CPC Class Codes: Y10S 707/99933 Query processing, i.e. searching
• CPC Class Codes: Y10S 707/99934 Query formulation, input preparation, or translation
• CPC Class Codes: Y10S 707/99935 Query augmenting and refining, e.g. inexact access
• CPC Class Codes: Y10S 707/99936 Pattern matching access
• CPC Class Codes: Y10S 707/99943 Generating database or data structure, e.g. via user interface
• CPC Class Codes: Y10S 707/99945 Object-oriented database structure processing
• CPC Class Codes: Y10S 707/99948 Application of database or data structure, e.g. distributed, multimedia, or image

Abstract

An implementation of a technology, described herein, for relevance-feedback, content-based facilitating accurate and efficient image retrieval minimizes the number of iterations for user feedback regarding the semantic relevance of exemplary images while maximizing the resulting relevance of each iteration. One technique for accomplishing this is to use a Bayesian classifier to treat positive and negative feedback examples with different strategies. In addition, query refinement techniques are applied to pinpoint the users'"'"' intended queries with respect to their feedbacks. These techniques further enhance the accuracy and usability of relevance feedback. This abstract itself is not intended to limit the scope of this patent. The scope of the present invention is pointed out in the appending claims.

Claims

1. A method for improving iterative results of Content-Based Image Retrieval (CBIR) using relevance feedback, the method comprising:

obtaining a set of positive feedback images and a set of negative feedback images via relevance feedback, the set of positive feedback images are those images deemed semantically relevant and the set of negative feedback images are those deemed semantically less relevant;

within a feature space, moving a positive candidate image towards the set of positive feedback images by adjusting distance metrics of the positive candidate image, the positive candidate images having similar low-level features as those of the set of positive feedback images, wherein the adjusting of distance metrics of the positive candidate image employs this evaluation;

$\mathrm{Dis}<br><br>\left(j\right)=\underset{i=0}{\overset{N_R}{\sum <br><br>}}<br><br>\text{\hspace{1em}<br><br>}<br><br>w_i*\left(1-\mathrm{Sim}<br><br>\left(j,X_i^{+}\right)\right);<br><br>$

 

where

j is the positive candidate image;

N_R is a number of images in the set of positive feedback images;

X_i⁺, i=1, . . . N_R is defined as the i^th image;

w_i is the normalized weight of the images in the set of positive feedback images;

within a feature space, distancing a negative candidate images from the set of positive feedback images by adjusting distance metrics of the negative candidate image, the negative candidate images having similar low-level features as those of the set of negative feedback images.

2. A method as recited in claim 1 further comprising constructing a Bayesian classifier of a positive feedback image by positive candidate images.

3. A method as recited in claim 1 further comprising employing a Bayesian decision boundary function to determine the probability of an image being a positive candidate image.

4. A method as recited in claim 1, wherein the moving further comprises:

normalizing features of an image within a feature space;

initializing parameters of the classifier;

updating the parameters using the features of the new positive feedback images;

calculating distances based upon a Bayesian decision boundary function;

sorting images based upon calculated distances, wherein the sort is performed as if no negative feedback images exist.

5. A computer comprising one or more computer-readable media having computer-executable instructions that, when executed by the computer, perform the method as recited in claim 1.

6. A computer-readable medium having computer-executable instructions that, when executed by a computer, performs the method as recited in claim 1.

7. A method for improving the evaluation of the similarity of images based upon image features, the method comprising calculating a weighted similarity distance between a first image and a set of images as a distance between them in a feature space, wherein the calculating employs this evaluation:

$\mathrm{Dis}<br><br>\left(j\right)=\underset{i=0}{\overset{N_R}{\sum <br><br>}}<br><br>\text{\hspace{1em}<br><br>}<br><br>w_i*\left(1-\mathrm{Sim}<br><br>\left(j,X_i^{+}\right)\right);<br><br>$

 

where

j is the first image;

N_R is a number of images in the set of images;

X_i⁺, i=1, . . . N_R is defined as the i^th image;

w_i is the normalized weight of the images in the set of images.

8. A computer comprising one or more computer-readable media having computer-executable instructions that, when executed by the computer, perform the method as recited in claim 7.

9. A computer-readable medium having computer-executable instructions that, when executed by a computer, performs the method as recited in claim 7.

10. A method for improving the evaluation of the similarity of images based upon image features, the method comprising defining a subject image to be similar to a set of example images because it is within a minimal distance of the set of example images within a feature space, wherein the distance between the subject image and the set of example images is calculated based upon this evaluation:

$\mathrm{Dis}<br><br>\left(j\right)=\underset{i=0}{\overset{N_R}{\sum <br><br>}}<br><br>\text{\hspace{1em}<br><br>}<br><br>w_i*\left(1-\mathrm{Sim}<br><br>\left(j,X_i^{+}\right)\right);<br><br>$

 

where

j is the subject image;

N_R is a number of images in the set of example images;

X_i⁺, i=1, . . . N_R is defined as the i^th image;

w_i is the normalized weight of the images in the set of example images.

11. A method as recited in claim 10 further comprising repeating the defining for each example image in a set of example images and for each subject image in a set of subject images.

12. A method as recited in claim 10 further comprising:

repeating the defining for each example image in a set of example images and for each subject image in a set of subject images;

ranking similarly of each subject image in the set of subject images to each example image in the set of example images.

13. A computer comprising one or more computer-readable media having computer-executable instructions that, when executed by the computer, perform the method as recited in claim 10.

14. A computer-readable medium having computer-executable instructions that, when executed by a computer, performs the method as recited in claim 10.

15. A system for improving iterative results of Content-Based Image Retrieval (CBIR) using relevance feedback, the system comprising:

a relevance feedback sub-system configured to obtain a set of positive feedback images and a set of negative feedback images via relevance feedback, the set of positive feedback images are those images deemed semantically relevant and the set of negative feedback images are those deemed semantically less relevant;

a feedback analyzer configured, within a feature space, to;

move a positive candidate images towards the set of positive feedback images by adjusting distance metrics of the positive candidate image, the positive candidate image having similar low-level features as those of the set of positive feedback images, wherein the adjusting of distance metric of the positive candidate image employs this evaluation;

$\mathrm{Dis}<br><br>\left(j\right)=\underset{i=0}{\overset{N_R}{\sum <br><br>}}<br><br>\text{\hspace{1em}<br><br>}<br><br>w_i*\left(1-\mathrm{Sim}<br><br>\left(j,X_i^{+}\right)\right);<br><br>$

 

where

j is the positive candidate image;

N_R is a number of images in the set of positive feedback images;

X_i⁺, i=1, . . . N_R is defined as the i^th image;

w_i is the normalized weight of the images in the set of positive feedback images;

distance a negative candidate image from the set of positive feedback images by adjusting distance metrics of the negative candidate images, the negative candidate image having similar low-level features as those of the set of negative feedback images.

16. A system as recited in claim 15, wherein the feedback analyzer is further configured to construct a Bayesian classifier of a positive feedback image by positive candidate images.

17. A system as recited in claim 15, wherein the feedback analyzer is further configured to define a subject image to be similar to an example image because it is within a minimal distance of the example image within a feature space.

18. A system for improving iterative results of Content-Based Image Retrieval (CBIR) using relevance feedback, the system comprising:

a memory comprising a set of computer program instructions; and

a processor coupled to the memory, the processor being configured to execute the computer program instructions, which comprise;

obtaining a set of positive feedback images and a set of negative feedback images via relevance feedback, the set of positive feedback images are those images deemed semantically relevant and the set of negative feedback images are those deemed semantically less relevant;

within a feature space, moving a positive candidate image towards the set of positive feedback images by adjusting distance metrics of the positive candidate image, the positive candidate image having similar low-level features as those of the set of positive feedback images, wherein the adjusting of distance metrics of the positive candidate image employs this evaluation;

$\mathrm{Dis}<br><br>\left(j\right)=\underset{i=0}{\overset{N_R}{\sum <br><br>}}<br><br>\text{\hspace{1em}<br><br>}<br><br>w_i*\left(1-\mathrm{Sim}<br><br>\left(j,X_i^{+}\right)\right);<br><br>$

 

where

j is the positive candidate image;

N_R is a number of images in the set of positive feedback images;

X_i⁺, i=1, . . . N_R is defined as the i^th image;

w_i is the normalized weight of the images in the set of positive feedback images;

within a feature space, distancing a negative candidate image from the set of positive feedback images by adjusting distance metrics of the negative candidate image, the negative candidate images having similar low-level features as those of the set of negative feedback images.

19. A system as recited in claim 18, wherein the properties of valid input indicate parameters of input by defining boundary delimitations of such parameters and define assumptions regarding such parameters.