A face alignment method based on SURF features

Abstract

Nowadays, face recognition research has been widely concerned, and facial face feature point positioning, that is, face alignment is an important part of the face recognition process, the accuracy of positioning and positioning speed directly affect the face recognition effect. The face alignment task in the real scene becomes a very difficult problem because of the presence of factors such as different pose, expression, illumination and partial occlusion in face images. Aiming at these problems, this paper presents a face alignment method based on SURF of Scale Invariant Feature Transform, which can quickly detect and characterize the key points of face image. In addition, We use a coarse to fine auto-encoder networks to implement complex non-linear mapping of face to face shape. Finally, By comparing the AFLW data set, It shows that the mean error rate of this method is 1.84%-2.74% lower than that of the traditional method, and It also has a good effect in the calculation speed.