Face detection and alignment method and device

Abstract

The invention discloses a face detection and alignment method and device, and the method comprises the steps: obtaining a face data set which comprises a plurality of face images with image label information; taking the face data set as input of a preset face detection and alignment network model to perform model training so as to output face label information predicted by the network; calculatinga first loss function corresponding to the face frame position information, a second loss function corresponding to the face frame classification information and a third loss function corresponding to the face key point coordinate information according to the image label information and the face label information; outputting a first weight factor corresponding to the first loss function, a secondweight factor corresponding to the second loss function and a third weight factor corresponding to the third loss function through a full connection layer of the network model; determining a total loss function according to each loss function and the corresponding weight factor; and according to the total loss function, performing back propagation to update each weight so as to determine a targetface detection and alignment network model.

Description

Technical field [0001] The present invention relates to the technical field of face recognition, and more specifically, to a method and device for face detection and alignment. Background technique [0002] In the field of face recognition, face detection and alignment is a key link in face recognition technology. The typical process of face recognition mainly includes three steps: [0003] The first step is face detection, which is to find out the position of all faces in a given image; [0004] The second step is face alignment, which is to correct the detected face; [0005] The third step is to perform feature extraction and feature comparison on the corrected face to complete the face recognition process. [0006] At present, there are mainly two face detection and alignment schemes in related technologies. One is that face detection and alignment use different networks for output, and the two use a serial structure. The deployment phase of this scheme requires more time. . The...

AI Technical Summary

Problems solved by technology

[0006] At present, in related technologies, there are mainly two face detection and alignment schemes, one of which is that face detection and alignment use different networks for output, and the two adopt a serial structure. The deployment phase of this scheme takes more time.

The other is that the face detection and alignment branches are trained in the same network. This solution focuses on the loss of frame regression and artificially increases its loss weight, resulting in inaccurate key point positioning and poor alignment, which affects face recognition.

Images