Authentication method and apparatus with transformation model

Patent metadata

Field	Value
Publication number	US-11688403-B2
Application number	US-202016811452-A
Country	US
Kind code	B2
Filing date	Mar 6, 2020
Priority date	Mar 15, 2019
Publication date	Jun 27, 2023
Grant date	Jun 27, 2023

Abstract

An authentication method and apparatus using a transformation model are disclosed. The authentication method includes generating, at a first apparatus, a first enrolled feature based on a first feature extractor, obtaining a second enrolled feature to which the first enrolled feature is transformed, determining an input feature by extracting a feature from input data with a second feature extractor different from the first feature extractor, and performing an authentication based on the second enrolled feature and the input feature.

First claim

What is claimed is:

1. A processor implemented authentication method, comprising:
obtaining a second enrolled feature, the second enrolled feature having been generated by transforming a first enrolled feature to the second enrolled feature, the first enrolled feature having been extracted from a first input data by a first feature extractor comprising a first neural network;
determining an input feature by extracting a feature from a second input data with a second feature extractor comprising a second neural network different from the first neural network; and
performing an authentication based on the second enrolled feature and the input feature.

2. The method of claim 1 , the transforming having been performed with a transformation model, the transformation model corresponding to a difference between the first neural network and the second neural network.

3. The method of claim 1 , wherein the second enrolled feature is received from an apparatus other than an apparatus that performs the authentication.

4. The method of claim 1 , wherein the second feature extractor is an updated version of the first feature extractor.

5. The method of claim 1 , wherein the second enrolled feature is obtained based on a transformation model.

6. The method of claim 5 , wherein the transformation model includes a structural element that corresponds to a structural difference between the first neural network and the second neural network.

7. The method of claim 5 , wherein the first neural network is pretrained to output first features for use with a first authentication procedure, wherein the second neural network is pretrained to output second features for a second authentication procedure, and wherein the transformation model is a neural network pretrained to transform features for use with the first authentication procedure to features for use with the second authentication procedure.

8. The method of claim 1 , wherein the first enrolled feature includes first sub features, and the second enrolled feature includes second sub-features to which the first sub features are respectively transformed.

9. The method of claim 8 , further comprising:
discarding at least a portion of the second sub features based on similarity measures of the second sub features with respect to each other.

10. The method of claim 9 , wherein the discarding comprises discarding at least one of the second sub features based on a similarity measure between the at least one of the second sub-features and an aggregate similarity of other of the second sub features.

11. The method of claim 10 , wherein the performing the authentication is based on a first threshold and a similarity between the second enrolled feature and the input feature, and
wherein the first threshold is equal to a second threshold based upon which the at least one of the second sub-features is discarded.

12. An authentication apparatus comprising:
one or more processors configured to:
obtain a second enrolled feature transformed from a first enrolled feature, wherein the first enrolled feature was extracted from an enrollment image based on a first feature extractor comprising a first neural network;
extract an input feature from an authentication image with a second feature extractor comprising a second neural network, wherein the first enrolled feature is transformed to the second enrolled feature based on a difference between the first neural network and the second neural network; and
perform an authentication based on the second enrolled feature and the input feature.

13. The apparatus of claim 12 , wherein the first enrolled feature is transformed to the second enrolled feature based on a transformation model that corresponds to the difference between the first neural network and the second neural network.

14. The apparatus of claim 13 , wherein the transformation model includes a structural element that is not a structural element of the first feature extractor and is a structural element of the second feature extractor.

15. The apparatus of claim 13 , wherein the first feature extractor neural network is pretrained to output first features for use with a first authentication procedure, wherein the second neural network is pretrained to output second features for a second authentication procedure, and wherein the transformation model is a neural network pretrained to transform features for use with the first authentication procedure to features for use with the second authentication procedure.

16. The apparatus of claim 12 , wherein the first enrolled feature includes first sub features, and the second enrolled feature includes second sub-features to which the first sub features are transformed.

17. The apparatus of claim 16 , wherein the one or more processors are configured to discard at least a portion of the second sub features based on a similarity between the second sub features.

18. The apparatus of claim 17 , wherein the one or more processors are configured to discard one of the second sub features based on a threshold and similarities between the one second sub feature and the remaining second sub features.

19. The apparatus of claim 12 , further comprising a memory storing instructions that, when executed by the one or more processors, configure the one or more processors to perform the extracting of the first enrolled feature, the obtaining of the second enrolled feature, the extracting of the input feature, and the performing of the authentication.

20. A method performed by a computing device, the method comprising:
extracting a first enrollment feature set from an enrollment image using a first feature extraction model, and storing the first enrollment feature set;
using the first feature extraction model to extract a first authentication feature set from a first authentication image and performing a first authentication based on the stored first enrollment feature set and the first authentication feature set;
using a second feature extraction model to extract a second feature set from a second authentication image;
transforming the stored first enrollment feature set to a second enrollment feature set based on a transformation model that corresponds to difference between the first feature extraction model and the second feature extraction model; and
performing a second authentication based on the second enrollment feature set and the second authentication feature set.

Assignees

• Samsung Electronics Co Ltd
• Snu R&Db Foundation

Inventors

• Han Seungju
• Han Jaejoon
• Ko Minsu
• Choi Chang Kyu
• Han Bohyung

Classifications

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/082

  modifying the architecture, e.g. adding, deleting or silencing nodes or connections · CPC title

• G06N3/09

  Supervised learning · CPC title

• G10L17/18

  Artificial neural networks; Connectionist approaches · CPC title

• G06V10/764

  using classification, e.g. of video objects · CPC title