Method and apparatus for facial recognition

Patent metadata

Field	Value
Publication number	US-10902245-B2
Application number	US-201816050436-A
Country	US
Kind code	B2
Filing date	Jul 31, 2018
Priority date	Sep 21, 2017
Publication date	Jan 26, 2021
Grant date	Jan 26, 2021

Abstract

Embodiments of the present disclosure disclose a method and apparatus for facial recognition. A specific embodiment of the method comprises: extracting a to-be-recognized dark light image captured in a dark light environment; inputting the dark light image into a pre-trained first convolutional neural network to obtain a target image after the dark light image is preprocessed, the first convolutional neural network being used to preprocess the dark light image; and inputting the target image into a pre-trained second convolutional neural network to obtain a facial recognition result, the second convolutional neural network being used to represent a corresponding relationship between the image and the facial recognition result. This embodiment improves accuracy of the facial recognition on the image captured in the dark light environment.

First claim

What is claimed is:

1. A method for facial recognition, comprising:
establishing a first convolutional neural network, wherein the establishing the first convolutional neural network comprises:
extracting a preset training sample, wherein the training sample includes a dark light sample image and a bright light sample image corresponding to the dark light sample image, and
using the dark light sample image as an input, outputting an image by the first convolutional neural network with the input, and training the first convolutional neural network based on a comparison between the outputted image of the first convolutional neural network and the bright light sample image corresponding to the inputted dark light sample image;

extracting a to-be-recognized dark light image captured in a dark light environment;
inputting the dark light image into the first convolutional neural network to obtain a target image after the dark light image is preprocessed, the first convolutional neural network being used to preprocess the dark light image; and
inputting the target image into a pre-trained second convolutional neural network to obtain a facial recognition result, the second convolutional neural network being used to represent a corresponding relationship between the image and the facial recognition result.

2. The method according to claim 1 , wherein the using the dark light sample image as an input, outputting an image by the first convolutional neural network with the input, and training the first convolutional neural network based on a comparison between the outputted image of the first convolutional neural network and the bright light sample image corresponding to the inputted dark light sample image comprises:
using the dark light sample image as the input, and training and obtaining the first convolutional neural network based on the bright light sample image and a preset loss function using a deep learning method, wherein a value of the loss function is used to represent a degree of difference between the image outputted by the first convolutional neural network and the bright light sample image.

3. The method according to claim 1 , wherein the using the dark light sample image as an input, outputting an image by the first convolutional neural network with the input, and training the first convolutional neural network based on a comparison between the outputted image of the first convolutional neural network and the bright light sample image corresponding to the inputted dark light sample image comprises:
extracting a pre-established generative adversarial network, wherein the generative adversarial network includes a generative network and a discriminative network, the generative network is a convolutional neural network for preprocessing an inputted image of the generative network, and the discriminative network is used to determine whether an inputted image of the discriminative network is an image outputted by the generative network;
using the dark light sample image as an input of the generative network, and using the image outputted by the generative network and the bright light sample image as the input of the discriminative network to obtain a discrimination result outputted by the discriminative network; and
statisticising an accuracy rate of the obtained discrimination result, and training, based on a machine learning method, the generative network and the discriminative network according to the accuracy rate, to define the generative network as the trained first convolutional neural network when the accuracy rate is a preset numerical value.

4. The method according to claim 1 , further comprising generating the training sample, wherein the generating the training sample comprises:
preprocessing a plurality of first bright light images pre-captured in a bright light environment, to obtain a first dark light image corresponding to each of the plurality of first bright light images; and
using the generated first dark light image as the dark light sample image, and using the plurality of first bright light images as the bright light sample image, to compose the training sample.

5. The method according to claim 1 , further comprising generating the training sample, wherein the generating the training sample comprises:
preprocessing a plurality of second dark light images pre-captured in the dark light environment to obtain a second bright light image corresponding to each of the plurality of second dark light images; and
using the plurality of second dark light images as the dark light sample image, and using the generated second bright light image as the bright light sample image, to compose the training sample.

6. The method according to claim 1 , wherein the dark light sample image is an image pre-captured in the dark light environment, and the bright light sample image corresponding to the dark light sample image is an image pre-captured in the bright light environment, wherein each dark light sample image and a corresponding bright light sample image are images of a given object captured at a given position and from a given angle.

7. An apparatus for facial recognition, comprising:
at least one processor; and
a memory storing instructions, the instructions when executed by the at least one processor, cause the at least one processor to perform operations, the operations comprising:
establishing a first convolutional neural network, wherein the establishing the first convolutional neural network comprises:
extracting a preset training sample, wherein the training sample includes a dark light sample image and a bright light sample image corresponding to the dark light sample image, and
using the dark light sample image as an input, outputting an image by the first convolutional neural network with the input, and training the first convolutional neural network based on a comparison between the outputted image of the first convolutional neural network and the bright light sample image corresponding to the inputted dark light sample image;

extracting a to-be-recognized dark light image captured in a dark light environment;
inputting the dark light image into the first convolutional neural network to obtain a target image after the dark light image is preprocessed, the first convolutional neural network being used to preprocess the dark light image; and
inputting the target image into a pre-trained second convolutional neural network to obtain a facial recognition result, the second convolutional neural network being used to represent a corresponding relationship between the image and the facial recognition result.

8. The apparatus according to claim 7 ,
wherein the using the dark light sample image as an input, outputting an image by the first convolutional neural network with the input, and training the first convolutional neural network based on a comparison between the outputted image of the first convolutional neural network and the bright light sample image corresponding to the inputted dark light sample image comprises:
using the dark light sample image as the input, and training and obtaining the first convolutional neural network based on the bright light sample image and a preset loss function using a deep learning method, wherein a value of the loss function is used to represent a degree of difference between the image outputted by the first convolutional neural network and the bright light sample image.

9. The apparatus according to claim 7 , wherein the using the dark light sample image as an input, outputting an image by the first convolutional neural network with the input, and training the first convolutional neural network based on a comparison between the outputted image of the first convoluti

Assignees

• Baidu online network technology beijing co ltd

Inventors

• Du Kang

Classifications

• G06V10/774

  Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06V10/82

  using neural networks · CPC title

• G06V10/764

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

• G06V40/172

  Classification, e.g. identification · CPC title

• G06N3/08 Primary

  Learning methods · CPC title