Method of extracting feature of input image based on example pyramid, and facial recognition apparatus

• US 9,875,397 B2
• Filed: 08/04/2015
• Issued: 01/23/2018
• Est. Priority Date: 09/16/2014
• Status: Active Grant

• Current Assignee: Samsung Electronics Co. Ltd.
• Original Assignee: Samsung Electronics Co. Ltd.
• Inventors: Hwang, Wonjun, Kim, Wonjun, Suh, Sungjoo, Kim, Jungbae, Han, Jaejoon
• Primary Examiner(s): Carter, Aaron W
• Application Number:
  US14/817,389
• Publication Number:
  US 20160078283A1
• Time in Patent Office: 903 Days
• CPC Class Codes: G06F 18/21345 enforcing sparsity or involving a domain transformation
• CPC Class Codes: G06F 18/24133 Distances to prototypes
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 10/7715 Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods
• CPC Class Codes: G06V 40/172 Classification, e.g. identification

Abstract

At least one example embodiment discloses a method of extracting a feature of an input image. The method includes constructing an example pyramid including at least one hierarchical level based on stored example images, generating a codebook in each of the at least one hierarchical level, calculating a similarity between the codebook and the input image, and extracting a feature of the input image based on the similarity.

Claims

1. A method of extracting a feature of an input image, the method comprising:

constructing an example pyramid including at least two hierarchical levels based on stored example images which include face images of different users;

generating a codebook in each of the at least two hierarchical levels;

calculating a similarity between each codebook and the input image; and

extracting a feature of the input image based on the similarity,

wherein the example pyramid is constructed using example image groups generated for each of the at least two hierarchical levels by clustering a feature vector of the stored example images based on a distance in a feature space.

2. The method of claim 1, wherein the constructing comprises:

generating the example image groups for each of the at least two hierarchical levels by clustering the stored example images based on a reference; and

constructing the example pyramid based on the example image groups.

3. The method of claim 2, wherein the generating the example image groups comprises:

projecting the feature vector of the stored example images to the feature space.

4. The method of claim 1, wherein the generating each codebook comprises:

generating a plurality of visual words based on the example image groups in each of the at least two hierarchical levels; and

generating each codebook based on the visual words.

5. The method of claim 4, wherein the generating the visual words comprises:

performing vector quantization on the example image groups in each of the at least two hierarchical levels; and

generating the visual words based on the vector quantization.

6. The method of claim 4, wherein the generating the visual words comprises:

performing sparse coding on the example image groups in each of the at least two hierarchical levels; and

generating the visual words based on the sparse coding.

7. The method of claim 1, wherein the calculating comprises:

measuring a plurality of distances between the input image and a plurality of groups associated with visual words in each codebook in the feature space; and

calculating the similarity based on the measured distances.

8. The method of claim 7, further comprising:

concatenating the measured distances, wherein the similarity is based on the concatenated measured distances.

9. The method of claim 1, wherein the extracting comprises:

extracting the feature of the input image based on a distribution of a probability value with respect to the similarity.

10. The method of claim 9, wherein the extracting comprises:

assigning a weight to the distribution of the probability value; and

extracting the feature of the input image based on the weight.

11. A non-transitory computer-readable medium comprising program code that, when executed by a processor, causes the processor to perform the method of claim 1.

20. The method of claim 8, further comprising:

adjusting a weight based on the concatenated measured distances; and

applying the adjusted weight to the measured distances, wherein the calculating the similarity calculates the similarity based on the applying.

21. The method of claim 20, further comprising:

determining a distribution of a probability value with respect to the similarity using the weight, wherein the extracting extracts the feature of the input image based on the distribution of the probability value with respect to the similarity.

12. A facial recognition apparatus comprising:

at least one processor configured to execute computer readable instructions to,

extract a facial area from an input image;

perform normalization on the facial area;

extract a feature of the input image based on the normalized facial area and a stored codebook;

recognize a face based on the extracted feature, the stored codebook is based on example image groups, the example image groups are in each level of an example pyramid using stored example images which include face images of different users; and

construct the example pyramid using the example image groups by clustering a feature vector of the stored example images based on a distance in a feature space; and

generate the stored codebook for the example image groups in each of at least two hierarchical levels of the example pyramid.

13. The apparatus of claim 12, wherein the at least one processor is configured to execute the computer readable instructions to calculate a similarity between the normalized facial area and the stored codebook, and extract the feature of the input image based on the similarity.

14. The apparatus of claim 13, wherein the at least one processor is configured to execute the computer readable instructions to measure a plurality of distances between the normalized facial area and a plurality of groups associated with visual words in the codebook in the feature space, and calculate the similarity based on the measured distances.

15. The apparatus of claim 14, wherein the at least one processor is configured to execute the computer readable instructions to assign a weight to a distribution of a probability value with respect to the similarity, and extract the feature of the input image based on the weight.

16. The apparatus of claim 12, wherein the at least one processor is configured to execute the computer readable instructions to recognize a face using a classifier based on the extracted feature.

17. The apparatus of claim 12, wherein the at least one processor is configured to execute the computer readable instructions to project the feature vector of the stored example images onto the feature space.

18. The apparatus of claim 12, wherein the at least one processor is configured to execute the computer readable instructions to generate a plurality of visual words based on the example image groups and generate the stored codebook based on the visual words.

19. The apparatus of claim 18, wherein at least one processor is configured to execute the computer readable instructions to perform vector quantization on the example image groups and generate the visual words based on the vector quantization.

22. The apparatus of claim 13, wherein the at least one processor is configured to execute the computer readable instructions to,

concatenate measured distances between the input image and visual words of the stored codebook,

adjust a weight based on the concatenated measured distances, and

apply the adjusted weight to the measured distances, wherein the processor is configured to calculate the similarity based on the applied adjusted weight.

23. The apparatus of claim 22, wherein the at least one processor is configured to execute the computer readable instructions to,

determine a distribution of a probability value with respect to the similarity using the weight, wherein the processor is configured to extract the feature of the input image based on the distribution of the probability value with respect to the similarity.