Method and apparatus for detecting human face

Patent metadata

Field	Value
Publication number	US-10936919-B2
Application number	US-201816052421-A
Country	US
Kind code	B2
Filing date	Aug 1, 2018
Priority date	Sep 21, 2017
Publication date	Mar 2, 2021
Grant date	Mar 2, 2021

Abstract

The present disclosure discloses a method and apparatus for detecting a human face. A specific embodiment of the method comprises: acquiring a to-be-detected image; inputting the to-be-detected image into a pre-trained first convolutional neural network to obtain facial feature information, the first convolutional neural network being used to extract a facial feature; inputting the to-be-detected image into a pre-trained second convolutional neural network to obtain semantic feature information, the second convolutional neural network being used to extract a semantic features of the image; and analyzing the facial feature information and the semantic feature information to generate a face detection result. This embodiment improves accuracy of a detection result of a blurred image.

First claim

What is claimed is:

1. A method for detecting a human face, comprising:
acquiring a to-be-detected image;
inputting the to-be-detected image into a pre-trained first convolutional neural network to obtain facial feature information, the first convolutional neural network being used to extract a facial feature;
inputting the to-be-detected image into a pre-trained second convolutional neural network to obtain semantic feature information, the second convolutional neural network being used to extract a semantic feature of the image, wherein the semantic feature information comprises: a probability that each point in the to-be-detected image belongs to a preset background category, a probability that each point in the to-be-detected image belongs to a preset hair category, a probability that each point in the to-be-detected image belongs to a preset eye category, a probability that each point in the to-be-detected image belongs to a preset nose category, a probability that each point in the to-be-detected image belongs to a preset mouth category, and a probability that each point in the to-be-detected image belongs to a preset skin color category; and
analyzing the facial feature information and the semantic feature information to generate a face detection result.

2. The method according to claim 1 , wherein the facial feature information comprises a first facial feature map and a plurality of second facial feature maps, wherein each point in the first facial feature map is used to represent a confidence level of the human face located in a region of the to-be-detected image corresponding to the each point of the first facial feature map, each point in each of the plurality of second facial feature maps is used to represent position information of a region of the to-be-detected image corresponding to the each point of the second facial feature map, and the first facial feature map and the plurality of second facial feature maps are respectively represented by matrixes.

3. The method according to claim 1 , wherein the semantic feature information is represented by a matrix.

4. The method according to claim 1 , wherein the analyzing the facial feature information and the semantic feature information to generate a face detection result comprises:
combining the facial feature information and the semantic feature information to generate combined feature information; and
inputting the combined feature information into a pre-trained third convolutional neural network to obtain the face detection result, wherein the third convolutional neural network is used to represent a correspondence of the face detection result to the facial feature information and the semantic feature information.

5. The method according to claim 2 , wherein the analyzing the facial feature information and the semantic feature information to generate a face detection result comprises:
combining the facial feature information and the semantic feature information to generate combined feature information; and
inputting the combined feature information into a pre-trained third convolutional neural network to obtain the face detection result, wherein the third convolutional neural network is used to represent a correspondence of the face detection result to the facial feature information and the semantic feature information.

6. The method according to claim 3 , wherein the analyzing the facial feature information and the semantic feature information to generate a face detection result comprises:
combining the facial feature information and the semantic feature information to generate combined feature information; and
inputting the combined feature information into a pre-trained third convolutional neural network to obtain the face detection result, wherein the third convolutional neural network is used to represent a correspondence of the face detection result to the facial feature information and the semantic feature information.

7. The method according to claim 4 , wherein the combining the facial feature information and the semantic feature information to generate combined feature information comprises:
combining the facial feature information and the semantic feature information to generate the combined feature information in any one of the following ways: multiplying the facial feature information by the semantic feature information based on corresponding elements, obtaining respective maximum values of respective corresponding elements of the facial feature information and the semantic feature information, and performing threshold truncation on the facial feature information and the semantic feature information based on the corresponding elements.

8. The method according to claim 4 , further comprising training the third convolutional neural network, comprising:
extracting a preset training sample, wherein the training sample comprises facial feature information and semantic feature information of a plurality of images;
combining facial feature information and semantic feature information of each image of the plurality of images to obtain combined feature information of the each image; and
using the combined feature information of the each image of the plurality of images as an input and the face detection result as an output to train and obtain the third convolutional neural network by using a machine learning method.

9. An apparatus for detecting a human face, 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:
acquiring a to-be-detected image;
inputting the to-be-detected image into a pre-trained first convolutional neural network to obtain facial feature information, the first convolutional neural network being used to extract a facial feature;
inputting the to-be-detected image into a pre-trained second convolutional neural network to obtain semantic feature information, the second convolutional neural network being used to extract a semantic feature of the image, wherein the semantic feature information comprises: a probability that each point in the to-be-detected image belongs to a preset background category, a probability that each point in the to-be-detected image belongs to a preset hair category, a probability that each point in the to-be-detected image belongs to a preset eye category, a probability that each point in the to-be-detected image belongs to a preset nose category, a probability that each point in the to-be-detected image belongs to a preset mouth category, and a probability that each point in the to-be-detected image belongs to a preset skin color category; and
analyzing the facial feature information and the semantic feature information to generate a face detection result.

10. The apparatus according to claim 9 , wherein the facial feature information includes a first facial feature map and a plurality of second facial feature maps, wherein each point in the first facial feature map is used to represent a confidence level of the human face located in a region of the to-be-detected image corresponding to the each point of the first facial feature map, each point in each of the plurality of second facial feature maps is used to represent position information of a region of the to-be-detected image corresponding to the each point of the second facial feature maps, and the first facial feature map and the plurality of second facial feature maps are respectively represented by matrixes.

11. The apparatus according to claim 9 , wherein the semantic feature information is represented by a matrix.

12. The apparatus accor

Assignees

• Baidu online network technology beijing co ltd

Inventors

• Du Kang

Classifications

• G06V10/806

  of extracted features · CPC title

• G06V10/82

  using neural networks · CPC title

• G06F18/253

  of extracted features · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/08 Primary

  Learning methods · CPC title