Image detection method, apparatus, device and storage medium

Patent metadata

Field	Value
Publication number	US-11244157-B2
Application number	US-202016879342-A
Country	US
Kind code	B2
Filing date	May 20, 2020
Priority date	Jul 5, 2019
Publication date	Feb 8, 2022
Grant date	Feb 8, 2022

Abstract

Example embodiments of the present disclosure provide an image detection method, an electronic device and a computer-readable storage medium. The image detection method includes the following. Am image to be detected including a target object is obtained and multiple feature representation determination modules are obtained. The multiple feature representation determination modules are trained for different parts of a reference object, using a reference image including the reference object and an authenticity of the reference image. Multiple feature representations for different parts of the target object are determined based on the image to be detected and the multiple feature representation determination modules. An authenticity of the image to be detected is determined based on the multiple feature representations.

First claim

What is claimed is:

1. A method for detecting an image, comprising:
acquiring an image to be detected comprising a target object and acquiring a plurality of feature representation determination models, wherein the plurality of feature representation determination models are trained for different parts of a reference object by using a reference image comprising the reference object and an authenticity of the reference image;
determining a plurality of feature representations for different parts of the target object based on the image to be detected and the plurality of feature representation determination models; and
determining an authenticity of the image to be detected based on the plurality of feature representations;
wherein determining the plurality of feature representations comprises:
determining a plurality of sets of key points associated with the different parts, from the image to be detected; and
determining the plurality of feature representations based on the plurality of sets of key points and the plurality of feature representation determination models;
wherein determining the plurality of feature representations based on the plurality of sets of key points and the plurality of feature representation determination models comprises:
determining an area in which one set of the plurality of sets of key points are located, from the image to be detected;
generating an intermediate image based on the area by highlighting the area and blurring other areas in the image to be detected; and
determining a feature representation of a part corresponding to the one set of key points by applying the intermediate image to the feature representation determination model corresponding to the one set of key points.

2. The method of claim 1 , wherein determining the authenticity of the image to be detected comprises:
determining an authenticity score for a part corresponding to one of the plurality of feature representations; and
in response to that the authenticity score is lower than a predetermined threshold, determining that the part corresponding to the one feature representation is generated by synthesis.

3. The method of claim 1 , further comprising:
generating at least one filtered image by applying at least one filter to the image to be detected; and
determining an additional feature representation for the image to be detected based on the at least one filtered image and an additional feature representation determination model, wherein the additional feature representation determination model is trained for the at least one filter, using the reference image and the authenticity of the reference image; and
wherein determining the authenticity of the image to be detected comprises:
determining a probability that the image to be detected is generated by synthesis based on the plurality of feature representations and the additional feature representation.

4. The method of claim 3 , wherein the at least one filter comprises a plurality of filters for different frequency ranges, and generating the at least one filtered image comprises:
applying one of the plurality of filters to the image to be detected, such that the filtered image generated comprises information of a frequency range corresponding to the one filter.

5. The method of claim 1 , wherein acquiring the plurality of feature representation determination models comprises:
generating a plurality of intermediate reference images corresponding to different parts of the reference object, based on the reference image; and
training the plurality of feature representation determination models, using the plurality of intermediate reference images and the authenticity of the reference image.

6. The method of claim 1 , wherein determining the authenticity of the image to be detected comprises:
determining the authenticity of the image to be detected by applying the plurality of feature representations to an authenticity evaluation model, wherein the authenticity evaluation model and the plurality of feature representation determination models are trained together, using the reference image and the authenticity of the reference image.

7. The method of claim 1 , further comprising:
enlarging the intermediate image to a specified size.

8. The method of claim 1 , wherein the highlighted area locates at center of the intermediate image.

9. An electronic device, comprising:
one or more processors; and
a storage device, configured to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors are caused to:
acquire an image to be detected comprising a target object and acquire a plurality of feature representation determination models, wherein the plurality of feature representation determination models are trained for different parts of a reference object by using a reference image comprising the reference object and an authenticity of the reference image;
determine a plurality of feature representations for different parts of the target object based on the image to be detected and the plurality of feature representation determination models; and
determine an authenticity of the image to be detected based on the plurality of feature representations;
wherein the one or more processors are caused to determine the plurality of feature representations by:
determining a plurality of sets of key points associated with the different parts, from the image to be detected; and
determining the plurality of feature representations based on the plurality of sets of key points and the plurality of feature representation determination models;
wherein the one or more processors are caused to determine the plurality of feature representations based on the plurality of sets of key points and the plurality of feature representation determination models by:
determining an area in which one set of the plurality of sets of key points are located, from the image to be detected;
generating an intermediate image based on the area; and
determining a feature representation of a part corresponding to the one set of key points by applying the intermediate image to the feature representation determination model corresponding to the one set of key points.

10. The electronic device of claim 9 , wherein the one or more processors are caused to determine the authenticity of the image to be detected by:
determining an authenticity score for a part corresponding to one of the plurality of feature representations; and
in response to that the authenticity score is lower than a predetermined threshold, determining that the part corresponding to the one feature representation is generated by synthesis.

11. The electronic device of claim 9 , wherein the one or more processors are caused to further:
generate at least one filtered image by applying at least one filter to the image to be detected; and determine an additional feature representation for the image to be detected based on the at least one filtered image and an additional feature representation determination model, wherein the additional feature representation determination model is trained for the at least one filter, using the reference image and the authenticity of the reference image; and
the one or more processors are caused to determine the authenticity of the image to be detected by:
determining a probability that the image to be detected is generated by synthesis based on the plurality of feature representations and the additional feature representation.

12. The electronic device of claim 11 , wherein at least one filter comprises a plurality of filters for different frequency ran

Assignees

• Baidu online network technology beijing co ltd

Inventors

• Xiong Junfeng
• Wang Yang
• Liu Yan
• Hao Xin

Classifications

• G06V10/764 Primary

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

• G06V40/18

  Eye characteristics, e.g. of the iris · CPC title

• G06V40/40

  Spoof detection, e.g. liveness detection · CPC title

• G06V10/82

  using neural networks · CPC title

• G06V20/80

  Recognising image objects characterised by unique random patterns · CPC title