Hierarchical face recognition training method and hierarchical face recognition method thereof

• US 8,306,282 B2
• Filed: 01/08/2009
• Issued: 11/06/2012
• Est. Priority Date: 09/19/2008
• Status: Active Grant

• Current Assignee: Altek Semiconductor Corp.
• Original Assignee: Altek Corporation
• Inventors: Chou, Hong-Long, Yang, Tai-Chang, Chang, Yin-Pin
• Primary Examiner(s): Akhavannik, Hadi
• Application Number:
  US12/350,378
• Publication Number:
  US 20100074479A1
• Time in Patent Office: 1,398 Days
• Field of Search: 382/118, 382/159
• US Class Current: 382/118
• CPC Class Codes: G06F 18/24323 Tree-organised classifiers
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 30/242 Division of the character sequences into groups prior to recognition; Selection of dictionaries
• CPC Class Codes: G06V 40/16 Human faces, e.g. facial parts, sketches or expressions

Abstract

A hierarchical face recognition training method and a hierarchical face recognition method thereof for performing a face feature recognition on an image under detection. The method includes a training process and a recognition process. The recognition method includes the steps. A plurality of training samples is obtained. The training samples are subdivided into a plurality of sub-image categories according to a plurality of angle intervals, and the training of a plurality of face features performs on a corresponding sub-image detector of each of the sub-image categories. The training measures performed repeatedly to generate sub-image categories at a sub-level of the sub-image categories. The training method includes the steps. An image under detection is loaded. A similarity of each of sub-image detectors compares according to the image under detection, and the sub-image detector having the highest similarity is selected. The face recognition measures performed repeatedly on the selected sub-image detector.

Claims

1. A hierarchical training method for face recognition, using a computer device to perform a face feature training of sub-image detectors at each level on a plurality of training samples, the method comprising:

obtaining the training samples;

performing a training measure by subdividing the training samples into a plurality of sub-image categories according to a plurality of angle intervals, and performing the face feature training on a corresponding sub-image detector of each of the sub-image categories; and

performing the training measure for each of the sub-image categories, so as to generate sub-image categories at a lower level than a current level for said each sub-image categories, through which one of the plurality of angle intervals corresponding to said each sub-image category is subdivided into a plurality of angle intervals at the lower level, each angle interval at the lower level corresponding to one of the plurality of sub-image categories at the lower level, and repeating the training measure for each sub-image category at the rawer level until a subdivision condition is satisfied, wherein

the sub-image categories of all levels form a tree structure.

2. The hierarchical training method for face recognition according to claim 1, after obtaining the training samples, further comprising:

performing a face recognition to recognize whether the training samples comprise a face portion;

performing the training measure if the training sample comprises the face portion; and

subdividing the training sample into an exceptional category if the training sample does not comprise the face portion.

3. The hierarchical training method for face recognition according to claim 1, wherein an entire face of the training sample is used as a face feature.

4. The hierarchical training method for face recognition according to claim 1, wherein a relative position of at least one of an eyebrow, an ear, an eye, a nose, and a mouth is selected from the face of the training sample as a face feature.

5. The hierarchical training method for face recognition according to claim 1, wherein the subdivision condition is determined by a number of generated levels of the sub-image categories.

6. The hierarchical training method for face recognition according to claim 1, wherein the subdivision condition is determined by a number of the sub-image categories at each level.

7. A hierarchical face recognition method performed on an image under detection by use of the sub-image detectors according to claim 1, using a computer device to perform a face feature recognition of each of the sub-image detectors on the image under detection, the method comprising:

loading the image under detection;

respectively performing the face recognition measure on the image under detection by each of the sub-image detectors at a same level, so as to obtain a similarity of each of the sub-image detectors to the image under detection, and selecting the sub-image detector having a highest similarity, from the sub-image detectors; and

repeatedly performing the face recognition measure on the image under detection by the selected sub-image detector until the sub-image detectors at a last level finish performing the face recognition measure on the image under detection.

8. The hierarchical face recognition method performed on the image under detection by use of sub-image detectors according to claim 7, after obtaining the similarity, further comprising:

performing the face recognition measure on the image under detection by the sub-image detector having the highest similarity; and

subdividing the image under detection into an exceptional category if a processing result of the face recognition measure satisfies a critical condition.

9. The hierarchical face recognition method performed on the image under detection by use of sub-image detectors according to claim 7, wherein when performing the face recognition measure, the sub-image detector selects at least one face feature as a basis for processing the similarity.

10. A hierarchical face recognition method for an image under detection, using a computer device to perform a face feature recognition of each of sub-image detectors on the image under detection, the method comprising:

obtaining a plurality of training samples;

performing a training measure by subdividing the training samples into a plurality of sub-image categories according to a plurality of angle intervals, and performing a face feature training on a corresponding sub-image detector of each of the sub-image categories;

performing the training measure for each of the sub-image categories, so as to generate sub-image categories at a lower level than a current level for said each sub-image categories, through which one of the plurality of angle intervals corresponding to said each sub-image category is subdivided into a plurality of angle intervals at the lower level, each angle interval at the lower level corresponding to one of the plurality of sub-image categories at the lower level, and repeating the training measure for each sub-image category at the lower level until a subdivision condition is satisfied, wherein the sub-image categories of all levels form a tree structure;

loading the image under detection;

performing a face recognition measure by comparing a similarity of each of the sub-image detectors at a same level as the image under detection according to the image under detection, and selecting the sub-image detector having a highest similarity from the same level; and

repeatedly performing the face recognition measure on the image under detection by the selected sub-image detector until the sub-image detectors at a last level finish performing the face recognition measure on the image under detection.

11. The hierarchical face recognition method for the image under detection according to claim 10, after obtaining the training samples, further comprising:

performing a face recognition to recognize whether the training samples comprise a face portion;

performing the training measure if the training sample comprises the face portion; and

subdividing the training sample into an exceptional category if the training sample does not comprise the face portion.

12. The hierarchical face recognition method for the image under detection according to claim 10, wherein an entire face of the training sample is used as a face feature.

13. The hierarchical face recognition method for the image under detection according to claim 10, wherein a relative position of at least one of an eyebrow, an ear, an eye, a nose, and a mouth is selected from the face of the training sample as a face feature.

14. The hierarchical face recognition method for the image under detection according to claim 10, wherein the subdivision condition is determined by a number of generated levels of the sub-image categories.

15. The hierarchical face recognition method for the image under detection according to claim 10, wherein the subdivision condition is determined by a number of the sub-image categories at each level.