Method, Apparatus and Computer Readable Recording Medium for Managing a Reference Face Database to Improve Face Recognition Performance Under a Restricted Memory Environment

• US 20140226877A1
• Filed: 09/27/2012
• Published: 08/14/2014
• Est. Priority Date: 09/27/2011
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Lee, Hyung Soo, Je, Hong Mo, Ryu, Woo Ju
• Granted Patent: US 9,298,976 B2
• Time in Patent Office: Days
• US Class Current: 382/118
• CPC Class Codes: G06F 16/51 Indexing; Data structures therefor; Storage structures
• CPC Class Codes: G06V 40/172 Classification, e.g. identification
• CPC Class Codes: G06V 40/50 Maintenance of biometric data or enrolment thereof

Abstract

Various embodiments of the present disclosure presents a method for managing a reference face database to improve face recognition performance under a restricted memory environment, said method comprising: acquiring a new input face image; determining a classification condition corresponding to the input face image with reference to classification conditions of existing registered face images of a reference face database used for face recognition; and if the classification condition corresponding to the input face image is a specific classification condition and the value of the face image corresponding to the specific classification condition in the reference face database is less than a preset threshold value, selecting the input face image as an image that can be additionally registered to the reference face database.

Claims

1. A method of managing a reference face database for improving performance of face recognition in a limited memory environment, the method comprising:

acquiring a new input facial image;

determining a classification condition relevant to the input facial image with reference to classification conditions of facial images previously registered in the reference face database used for face recognition; and

if the classification condition relevant to the input facial image is a specific classification condition and a facial image relevant to the specific classification condition is less sufficient than a preset threshold value in the reference face database, selecting the input facial image as an image that can be additionally registered in the reference face database.

2. The method of claim 1, wherein the classification condition of the facial images previously registered in the reference face database includes at least one of a facial expression of the input facial image, an angle of photographing the facial image, a direction of a light cast on the facial image, and a type of a light source.

3. The method of claim 1, wherein whether the facial image relevant to the specific classification condition is less sufficient than the preset threshold value in the reference face database is determined based on an identity of a person included in the input facial image.

4. The method of claim 1, further comprising selecting and deleting one or more of the facial images previously registered in the reference face database when the input facial image is additionally registered in the reference face database.

5. The method of claim 4, wherein the step of deleting comprises:

selecting the oldest facial image among the facial images previously registered in the reference face database; and

deleting the selected facial image.

6. The method of claim 4, wherein the step of deleting comprises:

counting the number of times each of the facial images previously registered in the reference face database were used for similarity matching, from a preset time point until present, to recognize an identity of a face included in the newly input image;

selecting an image having the smallest number of times used for the similarity matching among the facial images previously registered in the reference face database; and

deleting the selected facial image.

7. The method of claim 4, wherein the step of deleting comprises:

determining similarity between the facial images previously registered in the reference face database;

selecting facial images having the closest similarity between the previously registered facial images in the reference database;

further selecting one of the selected facial images; and

deleting the further selected facial image.

9. The method of claim 4, wherein the step of deleting comprises:

calculating a variance in a class between the facial images previously registered in the reference face database;

selecting a facial image having the smallest variance value; and

deleting the selected facial image.

10. The method of claim 9, wherein the step of calculating a variance comprises:

selecting one of the facial images previously registered in the reference face database;

calculating an average of face recognition matching values of the previously registered facial images other than the selected facial image; and

calculating a difference between a face recognition matching value of the one selected previously registered facial image and the average of the other face recognition matching values.

11. The method of claim 9, wherein the step of calculating a variance comprises:

calculating an average of face recognition matching values of the facial images previously registered in the reference face database; and

calculating a difference between the average of the face recognition matching values and a face recognition matching value of each of the previously registered facial image.

8. (canceled)

12. A method of managing a reference face database for improving performance of face recognition in a limited memory environment, the method comprising:

acquiring one or more new input facial images;

grouping the input facial images into groups of similar images by clustering the input facial images based on a specific face recognition metric;

selecting a facial image corresponding to a centroid facial image or a mean facial image among the grouped input facial images as a representative facial image; and

additionally registering the selected representative facial image in the reference face database used for face recognition.

13. The method of claim 12, wherein the step of grouping comprises:

calculating matching values for face recognition in the respective input facial images; and

grouping the input facial images into groups of highly similar images based on the calculated matching values.

14. The method of claim 13, wherein in the step of selecting a facial image as a representative facial image, a facial image having the matching value closest to a centroid matching value or mean matching value among the calculated matching values is selected as the representative facial image.

15. The method of claim 12, wherein the step of additionally registering the selected representative facial image in the reference face database comprises:

determining a classification condition relevant to the representative facial image with reference to classification conditions of the facial images previously registered in the reference face database; and

additionally registering the representative facial image in the reference face database, if the classification condition relevant to the representative facial image is a specific classification condition and a facial image relevant to the specific classification condition is less sufficient than a preset threshold value in the reference face database.

16-23. -23. (canceled)

24. A reference face database managing apparatus for improving performance of face recognition in a limited memory environment, the apparatus comprising:

a reference face database configured to record a facial image used for face recognition; and

an additional registration face selection unit configured to determine a classification condition relevant to a new input facial image with reference to classification conditions of facial images previously registered in the reference face database, if the input facial image is acquired, and configured to select the input facial image as an image that can be additionally registered in the reference face database, if the classification condition relevant to the input facial image is a specific classification condition and a facial image relevant to the specific classification condition is less sufficient than a preset threshold value in the reference face database.

25. The apparatus of claim 24, wherein the reference face database includes at least one of a facial expression of the input facial image, an angle of photographing the facial image, a direction of a light cast on the facial image, and a type of a light source, as the classification conditions of the previously registered facial images.

26. The apparatus of claim 24, wherein the additional registration face selection unit is configured to determine whether or not the facial image relevant to the specific classification condition is less sufficient than the preset threshold value in the reference face database based on an identity of a person included in the input facial image.

27. The apparatus of claim 24, further comprising a pre-registered face substitution unit configured to selecting and deleting some of the facial images previously registered in the reference face database when the input facial image is additionally registered in the reference face database.

28. The apparatus of claim 27, wherein the pre-registered face substitution unit is configured to select and delete the oldest facial image from among the facial images previously registered in the reference face database.

30. The apparatus of claim 27, wherein the pre-registered face substitution unit is configured to determine a similarity between the facial images previously registered in the reference face database, and select and delete facial images having the closest similarity between the previously registered facial images.

31. The apparatus of claim 30, wherein in determining the similarity, the pre-registered face substitution unit is configured to determine that similarity between two corresponding facial images is high as a difference between matching values is small, using the difference between the matching values for face recognition in the facial images previously registered in the reference face database.

32. The apparatus of claim 27, wherein the pre-registered face substitution unit is configured to calculate a variance in a class between the facial images previously registered in the reference face database, and select and delete a facial image having the smallest variance value.

29. (canceled)

33-34. -34. (canceled)

35. A reference face database managing apparatus for improving performance of face recognition in a limited memory environment, the apparatus comprising:

a reference face database configured to record a facial image used for face recognition; and

an additional registration face selection unit is configured to group one or more new input facial images into groups of similar images by clustering the input facial images based on a specific face recognition metric, if the input facial images are acquired, and is configured to select a facial image corresponding to a centroid facial image or a mean facial image among the grouped input facial images as a representative facial image, and additionally register the selected representative facial image in the reference face database.

36. The apparatus of claim 35, wherein the additional registration face selection unitis configured to calculate matching values for face recognition in the respective input facial images, and divide the input facial images into groups of highly similar images based on the calculated matching values.

37. The apparatus of claim 36, wherein the additional registration face selection unit is configured to select a facial image having the matching value closest to a centroid matching value or mean matching value among the calculated matching values as the representative facial image.

38. The apparatus of claim 35, wherein in additionally registering the selected representative facial image in the reference face database, the additional registration face selection unit determines a classification condition relevant to the representative facial image with reference to classification conditions of the facial images previously registered in the reference face database, and additionally registers the representative facial image in the reference face database, if the classification condition relevant to the representative facial image is a specific classification condition and a facial image relevant to the specific classification condition is less sufficient than a preset threshold value in the reference face database.

39. The apparatus of claim 35, further comprising a pre-registered face substitution unit for selecting and deleting one or more of the facial images previously registered in the reference face database when the input facial images are additionally registered in the reference face database.

40. The apparatus of claim 39, wherein the pre-registered face substitution unit is configured to count the number of times of using each of the facial images previously registered in the reference face database for similarity matching, from a preset time point until present to recognize an identity of a face included in the newly input image, and is configured to select and delete an image having the smallest number of times used for the similarity matching among the facial images previously registered in the reference face database.

41-42. -42. (canceled)