Face detection method

Abstract

The invention discloses a face detection method. The method comprises the following steps: 1, zooming an input image into different sizes according to a certain proportion through an image pyramid; then roughly predicting the coordinates of the human face, the confidence coefficient of the human face and the orientation of the human face through a first-level network in a sliding window manner, then ranking and filtering out most of negative samples according to the confidence coefficient, and sending the rest image blocks into a second-level network; 2, the second-level network further filtering out non-face samples and returns more accurate position coordinates, and giving a face orientation prediction result; 3, an angle arbitration mechanism carrying out final arbitration on the rotation angle of each sample by combining prediction results of the previous two networks; and 4, correcting each image block according to an arbitration result of an angle arbitration mechanism, and then sending the image blocks to a third-stage network for fine adjustment so as to predict the positions of the key points. According to the invention, the face with any rotation angle is aligned to the position of the standard face.

Description

technical field [0001] The invention relates to the technical field of face detection in the field of computer vision, in particular to a face detection method. Background technique [0002] Face detection has a wide range of applications in the fields of identity authentication, security, media and entertainment. The problem of face detection originated from face recognition and is a key step in realizing face recognition. The diversity of aspects such as , illumination, and scale brings great challenges to the detection of human faces and their key points. In the past ten years, a large number of methods have emerged in the field of computer vision to improve the ability of machines to detect human faces. The traditional Face detection methods can be divided into methods based on geometric features, methods based on skin color model and methods based on statistical theory according to the realization mechanism. Among them, the method based on geometric features mainly uses...

AI Technical Summary

Problems solved by technology

[0004] The paper by Sun Yi et al. (Sun Y , Wang X , Tang X . Deep Convolutional Network Cascade for Facial Point Detection[C] / / Computer Vision and Pattern Recognition, 2013 IEEE Conference on. IEEE, 2013.) introduced deep learning into human face for the first time Key point detection task, TCDCN (Zhang Z , Luo P , Loy C C , et al. Facial Landmark Detection by Deep Multi-task Learning[C] / / European Conference on Computer Vision.Springer, Cham, 2014.) Utilization and face key However, these methods are separated from face detection, which makes such methods highly dependent on the detection results of the previous step. HyperFace (Ranjan R, Patel V M, Chellappa R. HyperFace: A Deep Multi-task Learning Framework for Face Detection, Landmark Localization, PoseEstimation, and Gender Recognition[J]. IEEE Transactions on Pattern Analysis& Machine Intelligence, 2018, PP(99):1-1. ) Add more attribute labels to the training task, and improve the accuracy of key point regression through multi-task learning. However, too many learning tasks bring greater calculation and more running time. For face detection, this For a task with high real-time requirements, this method obviously has many limitations

Images