Classification and authentication of identification documents using a convolutional neural network

Patent metadata

Field	Value
Publication number	US-10410309-B2
Application number	US-201916286470-A
Country	US
Kind code	B2
Filing date	Feb 26, 2019
Priority date	Oct 17, 2016
Publication date	Sep 10, 2019
Grant date	Sep 10, 2019

Abstract

The present disclosure describes a method to determine whether a physical identification document is authentic. An authentication manager receives an image of a physical identification document to be authenticated. The authentication manager extracts a set of characteristics of the document from the image. The authentication manager determines a class of the document based on the set of characteristics. The authentication manager applies a convolution kernel, convolving the image with the kernel to generate a feature map. The authentication manager determines a score based on the feature map, where the score identifies the likelihood that the document is valid. The authentication manager then provides an indication that the physical document is authentic based on the score.

First claim

What is claimed is:

1. A method to determine whether physical identification documents are authentic, comprising:
receiving, by an authentication manager, an image of a physical identification document to be authenticated;
extracting, by the authentication manager, a set of characteristics of the physical identification document from the image;
determining, by the authentication manager, a class of the physical identification document based on the set of characteristics of the physical identification document;
applying, by the authentication manager and based on the class of the physical identification document, a convolution kernel to generate pixels of a feature map for a channel of the image by convolving the channel of the image with the convolution kernel;
determining, by the authentication manager, a score based on pixels of the feature map; and
providing, by the authentication manager, an indication that the physical identification document is authentic based on the score.

2. The method of claim 1 , wherein the image of the physical identification document comprises a plurality of channels.

3. The method of claim 2 , further comprising:
applying, by the authentication manager and based on the class of the physical identification document, a respective convolution kernel to generate pixels of a respective feature map for each channel of the image by convolving the respective channel of the image with the respective convolution kernel; and
determining, by the authentication manager, the score based on the respective feature map for each of the plurality of channels.

4. The method of claim 2 , further comprising:
generating, by the authentication manager, a respective feature map for each of the plurality of channels using the convolution kernel.

5. The method of claim 2 , wherein each of the channels corresponds to a respective color.

6. The method of claim 1 , further comprising:
training a convolutional neural network with a first plurality of scores from previously authenticated physical identification documents and a second plurality of scores from previously invalidated physical identification documents; and
determining the indication that the physical identification is authentic using the convolutional neural network.

7. The method of claim 1 , further comprising:
determining, by the authentication manager, a class of a first face of the physical identification document;
determining, by the authentication manager, a class of a second face of the physical identification document; and
determining, by the authentication manager, the score based on a comparison of the class of the first face of the physical identification document and the class of the second face of the physical identification document.

8. The method of claim 1 , further comprising:
normalizing, by the authentication manager, the image of the physical identification document.

9. The method of claim 8 , wherein normalizing the image of the physical identification document comprises at least one of removing a background from the image, rotating the image, deskewing the image, removing a glare from the image, correcting an exposure of the image, or correcting a blur of the image.

10. The method of claim 1 , further comprising:
determining, by the authentication manager, a subclass of the physical identification document based on the set of characteristics of the physical identification document; and
applying the convolution kernel based on both the class and the subclass of the physical identification document.

11. The method of claim 1 , wherein the set of characteristics of the physical identification document comprises a physical size of a barcode, a location of a barcode, or a location of a security feature.

12. A system to determine whether physical identification documents are authentic, the system comprising an authentication manger executable by one or more processors to:
receive an image of a physical identification document to be authenticated;
extract a set of characteristics of the physical identification document from the image;
determine a class of the physical identification document based on the set of characteristics of the physical identification document;
apply, based on the class of the physical identification document, a convolution kernel to generate pixels of a feature map for a channel of the image by convolving the channel of the image with the convolution kernel;
determine a score based on pixels of the feature map; and
provide an indication that the physical identification document is authentic based on the score.

13. The system of claim 12 , wherein image of the physical identification document comprises a plurality of channels.

14. The system of claim 13 , wherein the authentication manager is executable by the one or more processors to:
apply, based on the class of the physical identification document, a respective convolution kernel to generate pixels of a respective feature map for a respective channel of the image by convolving the respective channel of the image with the respective convolution kernel; and
determine the score based on the respective feature map for each of the plurality of channels.

15. The system of claim 13 , wherein the authentication manager is executable by the one or more processors to generate a respective feature map for each of the plurality of channels using the convolution kernel.

16. The system of claim 12 , wherein the authentication manager is executable by the one or more processors to:
train a convolutional neural network with a first plurality of scores from previously authenticated physical identification documents and a second plurality of scores from previously invalidated physical identification documents; and
determine the indication that the physical identification is authentic using the convolutional neural network.

17. The system of claim 12 , wherein the authentication manager is executable by the one or more processors to:
determine a class of a first face of the physical identification document;
determine a class of a second face of the physical identification document; and
determine the score based on a comparison of the class of the first face of the physical identification document and the class of the second face of the physical identification document.

18. The system of claim 12 , wherein the authentication manager is executable by the one or more processors to normalize the image of the physical identification document.

19. The system of claim 18 , wherein normalizing the image of the physical identification document comprises at least one of removing a background from the image, rotating the image, deskewing the image, removing a glare from the image, correcting an exposure of the image, or correcting a blur of the image.

20. The system of claim 12 , wherein the authentication manager is executable by the one or more processors to:
determine a subclass of the physical identification document based on the set of characteristics of the physical identification document; and
apply the convolution kernel based on both the class and the subclass of the physical identification document.

Assignees

• Facebook Inc

Inventors

• Rodriguez Raphael A
• Muesch Michael
• Thor David
• Boyd Justin
• Cahill Ryan
• Xiong Xiaolu

Classifications

• G06V10/82

  using neural networks · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06V10/243

  by compensating for image skew or non-uniform image deformations · CPC title

• G06V30/40

  Document-oriented image-based pattern recognition · CPC title

• G06V10/454

  Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN] · CPC title