Frontal face image synthesis method and system based on prior model

Abstract

The invention relates to a front face image synthesis method and system based on a prior model. The method includes the steps that firstly, sample standardization is conducted on a three-dimensional prototype sample library, a three-dimensional face sample library is acquired, and feature points of tested face images are calibrated; secondly, an average three-dimensional face model is acquired, and the attitude angles of the tested face images are acquired by establishing the corresponding relationship between projected feature points on the average three-dimensional face model and the calibrated feature points of the tested face images; thirdly, images more similar to the tested face images are screened out from the three-dimensional face sample library and used for establishing a prototype sample library; fourthly, front face images are synthesized. Under the condition that only information in the three-dimensional face library is adopted, the attitude angles of the input tested face images are estimated, corresponding front faces are synthesized, and certain detail features on original faces can be reserved. The method and system have theoretical and practical significance in research on face recognition and face attitude estimation and development of practical application.

Description

Technical field [0001] The invention relates to the fields of computer graphics, digital image processing and artificial intelligence, and in particular to a method and system for synthesizing a frontal face image through a face posture image. Background technique [0002] The frontal face synthesis methods include three-dimensional methods and two-dimensional methods. The idea of ​​the three-dimensional method is to perform three-dimensional reconstruction on the test face image. After obtaining the three-dimensional data of the test face image, the front face image can be obtained naturally. The most typical method of this type of method is the deformable model (3DMM, 3D Morphable Mpdel) method, which realizes the synthesis of face images from the side to the front, but the efficiency of the 3DMM face reconstruction process is very low. It is reconstructed on a server with a frequency of 2GHz A human face takes 300 seconds. Later scholars proposed many methods based on 3DMM i...

AI Technical Summary

Problems solved by technology

Some scholars first determine the pose angle of the test face, and then use the face training library of the corresponding pose to perform frontal face synthesis, but this synthesis method requires a large number of face training images of various poses

Images