Information processing apparatus capable of automatically setting degree of relevance between keywords, keyword attaching method and keyword auto-attaching apparatus

• US 6,480,841 B1
• Filed: 09/21/1998
• Issued: 11/12/2002
• Est. Priority Date: 09/22/1997
• Status: Expired due to Fees

• Current Assignee: Minolta Corporation Limited (Konica Minolta Inc.)
• Original Assignee: Minolta Corporation Limited (Konica Minolta Inc.)
• Inventors: Mochizuki, Takatoshi, Higashio, Kimihiko
• Primary Examiner(s): RONES, CHARLES
• Application Number:
  US09/157,413
• Time in Patent Office: 1,513 Days
• Field of Search: 707/3, 707/4, 707/529
• US Class Current: 1/1
• CPC Class Codes: G06F 16/51 Indexing; Data structures therefor; Storage structures
• 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 16/5838 using colour
• 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 retrieval apparatus allows keywords to be attached simply and automatically. A feature quantity extractor 15 extracts feature quantities of a registration-target image to obtain a p-dimensional feature vector. A vector converter 16 multiplies the p-dimensional feature vector with a canonical correlation coefficient matrix A2 so as to convert the feature vector into a k-dimensional feature composite variable vector, thus making a scale change. A keyword auto-setting processor 14 obtains q neighboring images based on distances between the feature composite variable vector of the registration-target image and coordinate composite variable vectors of r pre-registered images stored in a vector conversion coefficient storage 10. Then, based on distances between positions of the resulting q neighboring images on the image space and positions of the pre-registered keywords on the keyword space, h neighboring keywords are obtained for each of the neighboring images. Thus, q×h keywords suitably attached to the registration-target image are automatically set with simplicity.

Claims

1. An information processing apparatus in which an object is registered in storage with a keyword attached thereto, the apparatus comprising

object space defining means for, based on a correspondence between a plurality of objects and a plurality of keywords attached to these objects, defining an object space for the plurality of objects for subsequent retrieval of the object,

wherein the correspondence is obtained by mathematical quantification theory class III with each keyword used as a category and with each object used as a sample.

2. An information processing apparatus in which an object is registered in storage with a keyword attached thereto, the apparatus comprising

keyword space defining means for, based on a correspondence between a plurality of objects and a plurality of keywords attached to these objects, defining a keyword space for the plurality of keywords for subsequent retrieval of the object,

wherein the correspondence is obtained by mathematical quantification theory class III with each keyword used as a category and with each object used as a sample.

3. The information processing apparatus according to claim 2, further comprising

object space defining means for, based on the correspondence, defining an object space for the plurality of objects, wherein

the defined keyword space and object space are the same meaning space.

4. The information processing apparatus according to claim 3, further comprising

keyword registering means for, based on a position of an object the position of which on the object space is known and which is corresponded to a registration-target keyword, determining the position of the registration-target keyword on the keyword space, and placing the registration-target keyword at the position on the keyword space.

5. The information processing apparatus according to claim 3, further comprising:

keyword specifying means for specifying two keywords to be positioned at opposite poles; and

axis adding means for setting a new axis which interconnects the two specified keywords on the keyword space, and projecting all keywords already placed on the keyword space onto the set new axis to determine coordinates thereof on the new axis.

6. The information processing apparatus according to claim 2, further comprising:

keyword specifying means for specifying two keywords to be positioned at opposite poles; and

axis adding means for setting a new axis which interconnects the two specified keywords on the keyword space, and projecting all keywords already placed on the keyword space onto the set new axis to determine coordinates thereof on the new axis.

7. A keyword auto-attaching apparatus for automatically attaching a keyword to an object to be registered in storage, comprising:

meaning space storing means in which information as to a two- or more-dimensional meaning space in which a plurality of objects and keywords attached to these objects are placed is stored;

feature quantity extracting means for extracting feature quantities of a registration-target object;

object position calculating means for, based on the feature quantities, determining a position of the registration-target object on the meaning space;

keyword selecting means for selecting keywords positioned in a neighborhood of the position of the registration-target object out of the keywords placed on the meaning space; and

keyword attaching means for storing information that associates the selected keywords with the registration-target object for use in retrieving the object.

8. The keyword auto-attaching apparatus according to claim 7, further comprising

correlation calculating means for performing a canonical correlation analysis on positions on the meaning space and feature quantities relating to a plurality of objects, thereby determining a correlation between positions on the meaning space and feature quantities relating to the plurality of objects, wherein

the object position calculating means determines the position of the registration-target object on the meaning space based on the feature quantities of the registration-target object, by using the correlation.

9. The keyword auto-attaching apparatus according to claim 7, wherein

information as to the meaning space is divisionally stored in an object-use meaning space in which the plurality of objects are placed, and a keyword-use meaning space which has the same spatial structure as the object-oriented meaning space and in which the keywords attached to the plurality of objects are placed, and wherein

the keyword selecting means comprises:

neighboring object extracting means for extracting n (n;

integer) objects positioned in a neighborhood of the position of the registration-target object on the object-use meaning space; and

neighboring keyword extracting means for extracting keywords positioned in a neighborhood of the registration-target object based on the positions of the individual n extracted objects on the object-use meaning space, and by using the keyword-use meaning space.

10. The keyword auto-attaching apparatus according to claim 7, wherein

information as to the meaning space is divisionally stored in an object-use meaning space in which the plurality of objects are placed, and a keyword-use meaning space which has the same spatial structure as the object-use meaning space and in which the keywords attached to the plurality of objects are placed, and wherein

the keyword selecting means comprises:

neighboring object extracting means for extracting n objects positioned in a neighborhood of the position of the registration-target object on the object-use meaning space;

focused position calculating means for determining a focused position on the object-use meaning space based on the positions of the extracted n objects on the object-use meaning space; and

neighboring keyword extracting means for extracting keywords positioned in a neighborhood of the registration-target object based on the focused positions and by using the keyword-use meaning space.

11. A keyword attaching method for automatically attaching a keyword to an object to be registered in storage, comprising

preparing a two- or more-dimensional meaning space in which a plurality of objects and keywords attached to these objects are placed;

extracting feature quantities of a registration-target object;

determining a position of the registration-target object on the meaning space based on the feature quantities; and

selecting a keyword positioned in a neighborhood of the position of the registration-target object and storing information that associates the keyword with the registration-target object for subsequent retrieval of said object.

12. The keyword attaching method according to claim 11, further comprising

performing a canonical correlation analysis on positions on the meaning space and feature quantities relating to a plurality of objects, thereby determining a correlation between the positions on the meaning space and the feature quantities relating to the plurality of objects; and

determining the position of the registration-target object on the meaning space based on the feature quantities of the registration-target object by using the correlation.

13. The keyword attaching method according to claim 11, wherein

the meaning space is divisionally prepared in an object-use meaning space in which the plurality of objects are placed, and a keyword-use meaning space which has the same spatial structure as the object-use meaning space and in which the keywords attached to the plurality of objects are placed, the method further comprising;

selecting n (n;

integer) objects positioned in a neighborhood of the position of the registration-target object on the object-use meaning space; and

selecting a keyword positioned in a neighborhood of the registration-target object based on the positions of the individual n selected objects on the object-use meaning space and by using the keyword-use meaning space.

14. The keyword attaching method according to claim 11, wherein

the meaning space is divisionally prepared in an object-use meaning space in which the plurality of objects are placed, and a keyword-use meaning space which has the same spatial structure as the object-use meaning space and in which the keywords attached to the plurality of objects are placed, the method further comprising:

selecting n objects positioned in a neighborhood of the position of the registration-target object on the object-use meaning space;

determining a focused position on the object-use meaning space based on the positions of the selected n objects on the object-use meaning space; and

selecting a keyword positioned in a neighborhood of the registration-target object based on the focused positions and by using the keyword-use meaning space.

15. A computer program product in a memory which has stored therein a program for, based on a correspondence between a plurality of objects and a plurality of keywords attached to these objects, defining an object space for the plurality of objects for subsequent retrieval of at least one of the objects,

wherein the correspondence is obtained by mathematical quantification theory class III with each keyword used as a category and with each object used as a sample.

16. A computer program product in a memory, comprising a program for

extracting feature quantities of a registration-target object;

determining, based on the feature quantities, a position of the registration-target object on a two- or more-dimensional meaning space in which a plurality of objects and keywords attached to these objects are placed;

selecting keywords positioned in a neighborhood of the position of the registration-target object out of the keywords placed on the meaning space, and

storing information associating the keywords with the registration-target object for use in retrieving the object.

17. The computer program product in a memory according to claim 16, comprising a program for

performing a canonical correlation analysis on positions on the meaning space and feature quantities relating to a plurality of objects, thereby determining a correlation between positions on the meaning space and feature quantities relating to the plurality of objects; and

determining the position of the registration-target object on the meaning space based on the feature quantities of the registration-target object, by using the correlation.

18. A computer program product in a memory which has stored therein a program for, based on a correspondence between a plurality of objects and a plurality of keywords attached to these objects, defining a keyword space for the plurality of keywords for subsequent retrieval of at least one of the objects,

wherein the correspondence is obtained by mathematical quantification theory class III with each keyword used as a category and with each object used as a sample.

19. The computer program product in a memory according to claim 18, comprising

a program stored therein for defining an object space for the plurality of objects, which is the same meaning space as the defined keyword space, based on the correspondence.

20. The computer program product in a memory according to claim 19, comprising

a program stored therein for, based on a position of an object the position of which on the object space is known and which is corresponded to a registration-target keyword, determining the position of the registration-target keyword on the keyword space, and positioning the registration-target keyword at the position on the keyword space.

22. The computer program product in a memory according to claim 18, comprising

a program for setting on the keyword space a new axis which interconnects two specified keywords to be positioned at opposite poles with each other, and projecting all keywords already placed on the keyword space onto the set new axis to determine coordinates thereof on the new axis.

21. The computer program product in a memory according to claim 21, comprising

a program for setting on the keyword space a new axis which interconnects two specified keywords to be positioned at opposite poles with each other, and projecting all keywords already placed on the keyword space onto the set new axis to determine coordinates thereof on the new axis.

23. A method for automatically assigning keywords to data objects for using in retrieving said objects, comprising the steps of:

representing a plurality of data objects by means of respective n-dimensional object vectors in an n-dimensional space, where n>

1;

representing keywords associated with said data objects by means of respective n-dimensional keyword vectors in said space;

establishing correlation between said object vectors and said keyword vectors;

receiving information pertaining to a target data object;

determining an object vector in said space for said target object on the basis of the received information;

identifying at least one object vector within said space that is a neighbor of the object vector for said target object;

selecting at least one keyword vector that correlates with said identified object vector; and

storing information that associates the keyword corresponding to said selected keyword vector with said target object for use in retrieving said target object.

24. The method of claim 23, wherein the received information comprises feature quantities of said target object.

25. The method of claim 24, wherein the step of determining an object vector comprises the steps of:

representing said feature quantities as a p-dimensional feature vector;

transforming said p-dimensional feature vector into a first k-dimensional vector;

transforming said n-dimensional object vectors into second k-dimensional vectors; and

comparing said first k-dimensional vector with said second k-dimensional vectors to determine similarity between them.

26. The method of claim 23, wherein said correlation is established by mathematical quantification theory class III, with each keyword designated as a category and each data object designated as a sample.

27. The method of claim 23, further including the steps of:

specifying two keywords to be positioned at opposite poles of an axis;

defining an axis that interconnects the specified keywords within said space; and

projecting the keywords represented in said space onto said axis to determine coordinates for the keywords on said axis.

28. A computer-readable medium containing a program that executes the following steps:

representing a plurality of data objects by means of respective n-dimensional object vectors in an n-dimensional space, where n>

1;

representing keywords associated with said data objects by means of respective n-dimensional keyword vectors in said space;

establishing correlation between said object vectors and said keyword vectors;

receiving information pertaining to a target data object;

determining an object vector in said space for said target object on the basis of the received information;

identifying at least one object vector within said space that is a neighbor of the object vector for said target object;

selecting at least one keyword vector that correlates with said identified object vector; and

storing information that associates the keyword corresponding to said selected keyword vector with said target object for use in retrieving said target object.

29. The computer-readable medium of claim 28, wherein the received information comprises feature quantities of said target object.

30. The computer-readable medium of claim 29, wherein said program executes the following steps to determine the object vector for said target object:

representing said feature quantities as a p-dimensional feature vector;

transforming said p-dimensional feature vector into a first k-dimensional vector;

transforming said n-dimensional object vectors into second k-dimensional vectors; and

comparing said first k-dimensional vector with said second k-dimensional vectors to determine similarity between them.

31. The computer-readable medium of claim 28, wherein said correlation is established by mathematical quantification theory class III, with each keyword designated as a category and each data object designated as a sample.

32. The computer-readable medium of claim 28, wherein said program executes the following further steps:

receiving two keywords to be positioned at opposite poles of an axis;

defining an axis that interconnects the received keywords within said space; and

projecting the keywords represented in said space onto said axis to determine coordinates for the keywords on said axis.