Distance-space embedding for multi-descriptor matching and retrieval

1. A computerized process comprising:

receiving, at a computer processor, an archive of video data from which a plurality of archive descriptor types are extracted;

applying, using the computer processor, a query to the archive, the query comprising a number of N query descriptor types for a query object;

determining, using the computer processor, a difference between each query descriptor type and corresponding descriptor types of track segments in the archive, and storing the differences in one or more vectors;

storing each vector of differences in a computer storage medium as a point in an N dimensional space; and

identifying, using the computer processor, an archive object that is similar to the query object as a function of proximities of the differences to an origin in one or more dimensions of the N dimensional space.

2. The process of claim 1 , wherein the query descriptors are stored in and retrieved from a database.

3. The process of claim 1 , comprising:

identifying one or more archive objects as a function of a similarity of the query object and the archive objects;

receiving feedback from a user, the feedback comprising an indication of whether a particular archive object is similar to the query object; and

applying the received feedback to modify the query results.

4. The process of claim 1 , comprising applying a relevance weight to a descriptor type such that the descriptor type has more or less relevance as compared to other descriptor types in the identification of the archive object.

5. The process of claim 1 , comprising:

determining that one or more query descriptors are not discriminative;

expanding the query to include additional descriptor types; and

applying the expanded query to the archive.

6. The process of claim 5 , wherein the query is expanded by one or more of an automatic process and a process using feedback from a user.

7. The process of claim 1 , wherein only the set of archive objects that are similar to the query object as a function of at least a single distance are embedded in the N dimensional space.

8. A computerized process comprising:

receiving, at a computer processor, an archive of video data;

computing, using the computer processor, descriptors on the archive;

receiving, at the computer processor, a query video;

computing, using the computer processor, query descriptors on the query video;

determining, using the computer processor, a difference between each query descriptor type and corresponding descriptor types of track segments in the archive, and storing the differences in one or more vectors;

storing each vector of differences in a computer storage medium as a point in an N dimensional space;

retrieving a first set of descriptors with a high similarity to the query video for a particular descriptor type;

retrieving a second set of descriptors relating to a track segment for each descriptor in the first set of descriptors; and

measuring a similarity between a descriptor of the track segment and a query descriptor, the descriptor of the track segment and the query descriptor comprising the particular descriptor type.

9. The computerized process of claim 8 , comprising generating an N-dimensional point for each track segment based on N similarity measures.

10. The computerized process of claim 9 , wherein the similarity of track segments is measured from a proximity to the origin of the corresponding N-dimensional space.

11. The computerized process of claim 8 , comprising returning a set of the most similar tracks to a user.

12. The computerized process of claim 11 , comprising:

receiving feedback from a user; and

re-weighting dimensions in the N-dimensional space based on the feedback from the user.

13. A tangible computer readable storage device storing instructions, which, when executed by a processor, cause the processor to perform a process comprising:

receiving, at a computer processor, an archive of video data from which a plurality of archive descriptor types are extracted;

applying, using the computer processor, a query to the archive, the query comprising a number of N query descriptor types for a query object;

determining, using the computer processor, a difference between each query descriptor type and corresponding descriptor types of track segments in the archive, and storing the differences in one or more vectors;

storing each vector of differences in a computer storage medium as a point in an N dimensional space; and

identifying, using the computer processor, an archive object that is similar to the query object as a function of proximities of the differences to an origin in one or more dimensions of the N dimensional space.

14. The tangible computer readable storage device of claim 13 , wherein the query descriptors are stored in and retrieved from a database.

15. The tangible computer readable storage device of claim 13 , comprising instructions which when executed by a processor perform a process comprising:

identifying one or more archive objects as a function of a similarity of the query object and the archive objects;

receiving feedback from a user, the feedback comprising an indication of whether a particular archive object is similar to the query object; and

applying the received feedback to modify the query results.

16. The tangible computer readable storage device of claim 13 , comprising instructions which when executed by a processor perform a process comprising applying a relevance weight to a descriptor type such that the descriptor type has more or less relevance as compared to other descriptor types in the identification of the archive object.

17. The tangible computer readable storage device of claim 13 , comprising instructions which when executed by a processor perform a process comprising:

determining that one or more query descriptors are not discriminative;

expanding the query to include additional descriptor types; and

applying the expanded query to the archive.

18. The tangible computer readable storage device of claim 17 , wherein the query is expanded by one or more of an automatic process and a process using feedback from a user.

19. The tangible computer readable storage device of claim 13 , wherein only the set of archive objects that are similar to the query object as a function of at least a single distance are embedded in the N dimensional space.

20. The tangible computer readable storage device of claim 13 , comprising instructions which when executed by a processor perform a process comprising:

receiving feedback from a user; and

re-weighting dimensions in the N-dimensional space based on the feedback from the user.