Multi-angle face alignment method based on deep learning and system thereof and photographing terminal

Abstract

The invention discloses a multi-angle face alignment method based on deep learning and a system thereof and a photographing terminal. Marking of face key points and marking of face rotating angles are performed on a face sample image and the face sample image is inputted to a convolutional neural network to be trained, and different face pose types of the corresponding interval range of the face rotating angles are outputted so that face angle models of different face pose types are obtained; then regression training is performed on the face key point coordinates of the face sample image by utilizing the face angle models so that face alignment models corresponding to different face pose types are obtained; and finally an image to be detected is inputted to the face angle models to perform face angle detection, and the face alignment model of the corresponding angle is called to perform regression prediction. Precision is high, robustness is high and space occupation of the models obtained through training is low, and the method is particularly suitable for the face alignment application of which the situations of the photographed person are complex, the requirement for precision is high and occupation of the algorithm in the physical space is required to be low.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a multi-angle face alignment method based on deep learning and a system and a shooting terminal for applying the method. Background technique [0002] The main purpose of face alignment is to allow the computer to automatically locate the positions of various parts of the face, such as the corners of the eyes, the corners of the mouth, and the tip of the nose. In the multi-angle rotation, in order to reduce the difficulty of recognition, the face of the person being photographed is turned to the same angle through face alignment to improve the accuracy of face recognition; the second is the makeup of the picture. In order to perform virtual makeup on the face in the picture, it is necessary The position of each component of the face is located by the face alignment method. [0003] In the existing face alignment methods, the key point position data is first obtained by ...

AI Technical Summary

Problems solved by technology

However, in daily shooting scenes, the subject is not always facing the camera. In many cases, the eyes and other parts of the face will be blocked due to different angles of the side face, resulting in unsatisfactory face alignment, and the accuracy is also high. Less than the demand for makeup on pictures, the trained model takes up a lot of space and is difficult to embed into mobile devices such as mobile phones

Images