AUTHENTICATION METHOD AND APPARATUS, AND METHOD AND APPARATUS FOR TRAINING A RECOGNIZER

• US 20170124428A1
• Filed: 10/18/2016
• Published: 05/04/2017
• Est. Priority Date: 11/04/2015
• Status: Active Grant

• Current Assignee: Samsung Electronics Co. Ltd.
• Original Assignee: Samsung Electronics Co. Ltd.
• Inventors: HAN, SEUNGJU, YOO, BYUNGIN, HAN, JAEJOON
• Granted Patent: US 10,275,684 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 18/22 Matching criteria, e.g. proximity measures
• CPC Class Codes: G06F 18/285 Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system
• CPC Class Codes: G06F 21/32 using biometric data, e.g. fingerprints, iris scans or voiceprints
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06V 10/454 Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
• CPC Class Codes: G06V 10/761 Proximity, similarity or dissimilarity measures
• CPC Class Codes: G06V 10/82 using neural networks
• CPC Class Codes: G06V 40/172 Classification, e.g. identification

Abstract

An authentication method and apparatus, and a method and apparatus for training a recognizer are provided. The authentication method may include determining whether authentication of an input image succeeds by comparing a similarity associated with an intermediate layer to a threshold.

Claims

1. An authentication method comprising:

applying an input image to a recognizer, the recognizer including a plurality of layers; and

performing authentication of the input image based on at least one feature value of the input image and at least one feature value of a registered image corresponding to the at least one intermediate layer, the at least one feature value of the input image output by at least one intermediate layer of the layers in the recognizer.

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

determining whether the input image is e accepted or rejected.

3. The authentication method of claim 1, wherein the plurality of layers are layers of a deep learning (DL) network.

4. The authentication method of claim 1, wherein the at least one intermediate layer is a plurality of intermediate layers, and the performing comprises:

determining whether the authentication of the input image succeeds based on feature values of the input image and feature values of the registered image, the feature values of the input image output by the intermediate layers, respectively, and the feature values of the registered image corresponding to the intermediate layers, respectively.

5. The authentication method of claim 1, wherein the performing comprises:

determining a similarity between the at least one feature value of the input image and the at least one feature value of the registered image; and

performing the authentication based on the similarity to a threshold.

6. The authentication method of claim 5, wherein the performing comprises:

determining that the input image is accepted when the determined similarity exceeds a similarity corresponding to a false acceptance rate (FAR) of the intermediate layer.

7. The authentication method of claim 6, wherein the performing comprises:

determining that the input image is rejected when the determined similarity is less than a similarity corresponding to a verification rate (VR) of the intermediate layer.

8. The authentication method of claim 7, wherein the performing comprises:

proceeding to a next layer of the recognizer when the determining similarity exceeds the similarity corresponding to the VR and is less than the similarity corresponding to the FAR.

9. The authentication method of claim 5, wherein the threshold comprises a first threshold part based on a false acceptance rate (FAR) and a second threshold part based on a verification rate (VR).

10. The authentication method of claim 9, wherein the first threshold part is greater than the second threshold part.

11. The authentication method of claim 1, wherein the performing comprises:

determining a distance between the at least one feature value of the input image and the at least one feature value of the registered image; and

performing the authentication based on the distance threshold.

12. The authentication method of claim 11, wherein the performing comprises:

determining that the input image is accepted when the determined distance is less than a distance corresponding to a false acceptance rate (FAR) of the intermediate layer.

13. The authentication method of claim 12, wherein the performing comprises:

determining that the input image is rejected when the distance exceeds a distance corresponding to a verification rate (VR) of the intermediate layer.

14. The authentication method of claim 13, wherein the performing comprises:

proceeding to a next layer of the recognizer when the determined distance exceeds the distance corresponding to the FAR and is less than the distance corresponding to the VR.

15. The authentication method of claim 11, wherein the threshold comprises a first threshold part based on a false acceptance rate (FAR) and a second threshold part based on a verification rate (VR).

16. The authentication method of claim 15, wherein the first threshold part is greater than the second threshold part.

17. The authentication method of claim 1, wherein the performing further includes:

performing the authentication of the input image based on a final layer feature value of the input image and a final layer feature value of the registered image, the final layer feature value output by a final layer in the recognizer and the final layer feature value of the registered image corresponding to the final layer.

18. The authentication method of claim 1, further comprising:

applying the registered image to the recognizer; and

storing the at least one feature value output by the at least one intermediate layer and a final layer feature value output by a final layer of the plurality of layers as feature values of the registered image.

19. The authentication method of claim 18, further comprising:

receiving the registered image from a camera.

20. The authentication method of claim 1, wherein the at least one intermediate layer comprises a plurality of intermediate layers among the plurality of layers.

21. The authentication method of claim 1, wherein the plurality of layers comprise at least one convolution layer and at least one pooling layer.

22. A training method comprising:

training a recognizer, the recognizer including a plurality of layers;

determining a first threshold for an intermediate layer in the trained recognizer based on a desired performance index; and

determining a second threshold for a final layer in the trained recognizer based on the desired performance index.

23. The training method of claim 22, further comprising:

applying validation data to the trained recognizer;

determining a first similarity between intermediate feature values, the intermediate feature values output by the intermediate layer;

determining a second similarity between final feature values, the final feature values output by the final layer;

determining the first threshold based on the first similarity; and

determining the second threshold based on the second similarity.

24. The training method of claim 22, further comprising:

applying validation data to the trained recognizer;

determining a first distance between intermediate feature values, the intermediate feature values output by the intermediate layer;

determining a second distance between final feature values, the final feature values output by the final layer;

determining the first threshold based on the first distance; and

determining the second threshold based on the second distance.

25. The training method of claim 22, wherein the first threshold and the second threshold are used to determine whether an input image is accepted or rejected.

26. The training method of claim 22, wherein the determining of the first threshold comprises:

determining the first threshold based on a similarity between a first intermediate feature value of the intermediate feature values and a second intermediate feature value of the intermediate feature values.

27. The training method of claim 22, wherein the determining of the first threshold comprises:

determining the first threshold based on a distance between a first intermediate feature value of the intermediate feature values and a second intermediate feature value of the intermediate feature values.

28. The training method of claim 22, wherein the training of the recognizer comprises:

training the recognizer based on an output of the intermediate layer, an output of the final layer and a label value of the training sample, the output of the intermediate layer being in response to an input of a training sample, the output of the final layer being in response to the input of the training sample.

29. The training method of claim 22, wherein the training of the recognizer comprises:

training the recognizer based on an intermediate feature value output by a first network a final feature value and a label of a training sample, the first network receives an output of the intermediate layer, the final feature value output by a second network that receives an output of the final layer.

30. An authentication apparatus comprising:

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

output at least one intermediate feature value of an input image using at least one intermediate layer among a plurality of layers included in a recognizer, and

perform authentication of the input image based on the at least one intermediate feature value of the input image output using the intermediate layer and a feature value of a registered image corresponding to the intermediate layer.

31. The authentication apparatus of claim 30, wherein the at least one processor is further configured to execute the computer-readable instructions to determine whether to accept the input image or reject the input image.

32. The authentication apparatus of claim 30, wherein the plurality of layers are layers of a deep learning (DL) network.

33. The authentication apparatus of claim 30, wherein the at least one intermediate layer comprises at least one layer other than a final layer among the plurality of layers.

34. The authentication apparatus of claim 30, wherein the at least one intermediate layer is a plurality of intermediate layers, and

the at least one processor is further configured to execute the computer-readable instructions to determine whether the authentication of the input image succeeds based on intermediate feature values of the input image and feature values of the registered image, the feature values of the input image output by the intermediate layers, respectively, and the feature values of the registered image corresponding to the intermediate layers, respectively.

35. The authentication apparatus of claim 30, wherein the at least one processor is further configured to execute the computer-readable instructions to

determine a similarity between the at least one feature value of the input image and the at least one feature value of the registered image, and

perform the authentication based on the similarity and a threshold.

36. The authentication apparatus of claim 30, wherein the at least one processor is further configured to execute the computer-readable instructions to,

determine a distance between the at least one feature value of the input image and the at least one feature value of the registered image, and

perform the authentication based on the distance and a threshold.

37. The authentication apparatus of claim 30, wherein the at least one processor is further configured to execute the computer-readable instructions to apply the registered image to the recognizer and to store the at least one feature value output by the at least one intermediate layer and a final layer feature value output by a final layer as feature values of the registered image.