Semi-automatic annotation of multimedia objects

• US 20050114325A1
• Filed: 10/20/2004
• Published: 05/26/2005
• Est. Priority Date: 10/30/2000
• Status: Active Grant

• Current Assignee: SZ DJI Technology Co., Ltd. (d/b/a DJI) (iFlight Technology Company Limited)
• Original Assignee: Microsoft Corporation
• Inventors: Zhang, Hong-Jiang, Liu, Wen-Yin
• Granted Patent: US 7,349,895 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/48 Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
• CPC Class Codes: G06F 16/58 Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
• CPC Class Codes: G06F 16/583 using metadata automatically derived from the content
• CPC Class Codes: Y10S 707/99933 Query processing, i.e. searching
• CPC Class Codes: Y10S 707/99934 Query formulation, input preparation, or translation

Abstract

A multimedia object retrieval and annotation system integrates an annotation process with object retrieval and relevance feedback processes. The annotation process annotates multimedia objects, such as digital images, with semantically relevant keywords. The annotation process is performed in background, hidden from the user, as the user conducts normal searches. The annotation process is “semi-automatic” in that it utilizes both keyword-based information retrieval and content-based image retrieval techniques to automatically search for multimedia objects, and then encourages users to provide feedback on the retrieved objects. The user identifies objects as either relevant or irrelevant to the query keywords and based on this feedback, the system automatically annotates the objects with semantically relevant keywords and/or updates associations between the keywords and objects. As the retrieval-feedback-annotation cycle is repeated, the annotation coverage and accuracy of future searches continues to improve.

Claims

1. A computer-readable medium having computer-executable instructions for performing steps comprising:

identifying, in response to a search query, first multimedia objects having an associated keyword that matches a keyword in the search query, and identifying second multimedia objects that have content features similar to those of the first multimedia objects, wherein the content features do not match the keyword in the search query;

presenting the first and second multimedia objects to a user;

monitoring feedback from the user as to which of the first and second multimedia objects are relevant to the search query;

maintaining associations between the keywords and the multimedia objects, the associations being weighted to indicate how relevant the keywords are to the multimedia objects;

adjusting the weights of the associations based on the user'"'"'s feedback, wherein the adjusting comprises increasing a weight of an association between the keyword and a particular multimedia object that is deemed relevant by the user or/and decreasing a weight of an association between the keyword and a particular multimedia object that is deemed irrelevant by the user; and

annotating one or more of the second multimedia objects, which are deemed relevant by the user, with the keyword.

2. The computer readable medium of claim 1 having further computer-readable instructions for performing the step of removing the keyword from the particular multimedia object in an event that the weight is less than a threshold value.

3. The computer readable medium of claim 1 having further computer-readable instructions for performing the step of training how the first and second multimedia objects are identified based on the user'"'"'s feedback.

4. The computer readable medium of claim 1 having further computer-readable instructions for performing the step of refining the search to identify additional multimedia objects that contain content features similar to those of the multimedia objects indicated by the user as being relevant.

5. The computer readable medium of claim 1 wherein the multimedia objects comprise one of digital images, video objects, and audio objects.

6. A computer-readable medium having computer-executable instructions for performing steps comprising:

identifying, in response to a search query, first multimedia objects having an associated keyword that matches a keyword in the search query, and identifying second multimedia objects that have content features similar to those of the first multimedia objects, wherein the content features do not match the keyword in the search query;

presenting the first and second multimedia objects to a user;

monitoring feedback from the user as to which of the first and second multimedia

objects are relevant to the search query;

maintaining associations between the keywords and the multimedia objects, the associations being weighted to indicate how relevant the keywords are to the multimedia objects;

adjusting the weights of the associations based on the user'"'"'s feedback, wherein the adjusting comprises increasing a weight of an association between the keyword and a particular multimedia object that is deemed relevant by the user or/and decreasing a weight of an association between the keyword and a particular multimedia object that is deemed irrelevant by the user; and

annotating one or more of the second multimedia objects, which are deemed relevant by the user, with the keyword.

7. The computer readable medium of claim 6 wherein the step of presenting comprises using content-based information retrieval to retrieve the multimedia objects.

8. The computer readable medium of claim 6 wherein the step of presenting comprises using both content-based information retrieval and semantic-based information retrieval to retrieve the multimedia objects.

9. The computer readable medium of claim 6 wherein the step of monitoring comprises monitoring both feature-based relevance feedback and semantic-based relevance feedback.

10. The computer readable medium of claim 6 wherein the step of annotating is hidden from the user.

11. The computer readable medium of claim 6 wherein the step of annotating comprises:

in an event that a particular multimedia object is deemed irrelevant by the user and is already annotated with the keyword, weakening an association between the keyword and the particular multimedia object.

12. A computer-readable medium having computer-executable instructions for performing steps comprising:

identifying, in response to a search query, first multimedia objects having an associated keyword that matches a keyword in the search query, and identifying second multimedia objects that have content features similar to those of the first multimedia objects,

wherein the content features do not match the keyword in the search query;

presenting the first and second multimedia objects to a user;

monitoring feedback from the user as to which of the multimedia objects are relevant;

annotating one or more of the multimedia objects based on the user'"'"'s feedback, with a keyword;

in an event that a particular multimedia object is deemed relevant by the user and not yet annotated with the keyword, adding the keyword to the particular multimedia object; and

in an event that the particular multimedia object is deemed relevant by the user and is already annotated with the keyword, strengthening an association between the keyword and the particular multimedia object.

13. The computer readable medium of claim 12 wherein the step of monitoring comprises monitoring both feature-based relevance feedback and semantic-based relevance feedback.

14. The computer readable medium of claim 12 wherein the step of annotating is hidden from the user.

15. The computer readable medium of claim 12 wherein the step of annotating comprises:

in an event that a particular multimedia object is deemed irrelevant by the user and is already annotated with the keyword, weakening an association between the keyword and the particular multimedia object.

16. The computer readable medium of claim 12 wherein the step of annotating comprises:

in an event that a particular multimedia object is deemed irrelevant by the user and is already annotated with the keyword, removing the keyword from the particular multimedia object.

17. An image retrieval system comprising:

a means for retrieving multimedia objects from a database based on a search query first multimedia objects having an associated keyword that matches a keyword in the search query, and identifying second multimedia objects that have content features similar to those of the first multimedia objects wherein the content features do not match the keyword in the search query;

a means for capturing a user'"'"'s feedback as to whether the multimedia objects are relevant to the search query; and

a means for annotating, with the keyword, the second multimedia objects based on the user'"'"'s feedback,

wherein the search query comprises a keyword-based search query having at least one keyword and in an event that a particular multimedia object is deemed relevant by the user and is already annotated with the keyword, the annotation unit strengthens an association between the keyword and the particular multimedia object, and/or in an event that a particular multimedia object is deemed irrelevant by the user and is already annotated with the keyword, removing the keyword from the particular multimedia object, and/or in an event that a particular multimedia object is deemed irrelevant by the user and is already annotated with the keyword, weakening an association between the keyword and the particular multimedia object.

18. The image retrieval system as recited in claim 17, wherein the search query comprises a keyword-based search query having one or more keywords.

19. The image retrieval system as recited in claim 17, wherein the search query comprises a content-based search query having one or more content features.

20. The image retrieval system as recited in claim 17, wherein the means for retrieving information employs both content-based information retrieval and semantic-based information retrieval.

21. The image retrieval system as recited in claim 17, wherein the means for retrieving information comprises:

a means for handling both keyword-based queries having one or more search keywords and content-based queries having one or more content features of a multimedia object; and

a means for identifying at least one of (1) first multimedia objects having keywords that match the search keywords from a keyword-based query, and (2) second multimedia objects having content features similar to the content features of a content-based query.

22. The image retrieval system as recited in claim 17, wherein the means for capturing employs both feature-based relevance feedback and semantic-based relevance feedback.

23. The image retrieval system as recited in claim 17, wherein:

the search query comprises a keyword-based search query having at least one keyword; and

in an event that a particular multimedia object is deemed relevant by the user and is not yet annotated with the keyword, the annotation unit adds the keyword to the particular multimedia object.

24. The image retrieval system as recited in claim 17, wherein the means for capturing comprises a feedback analyzer to train the system based on the user'"'"'s feedback.