Research On Natural Evolution Of Human Face Based On Semi-supervised Adversarialauto-encoder

Face image is the most common data.In recent years,with the development of deep learning,the research on the natural evolution of human face has become increasingly hot.The natural evolution of the face refers to the generation of face images of other age groups of the person according to a person's face image of a certain age group,presenting a face image with “regression” or “aging” effect,while still retaining personal personality Features(ie,identity information).This paper takes the natural evolution of human face as the research goal.Based on the full study of previous methods,the face natural evolution algorithm(CAAE+ algorithm)based on semi-supervised adversarial auto-encoder is proposed.The main work of the thesis includes the following three aspects:Face image preprocessing.Most of the pictures in the face image database are pictures taken by life or by the Internet.Since the positions of the faces in the pictures are different,the face images are pre-processed to highlight the face parts to obtain the same standard face images.In this paper,AdaBoost algorithm is used to locate the position of human face and the human eye relative to the whole image.On the basis of this,the face image is normalized by rotation and scaling.A face natural evolution algorithm(CAAE+ algorithm)based on semi-supervised adversarial auto-encoder is proposed.After researching the generative adversarial networks,the semi-supervised adversarial auto-encoder and the conditional adversarial auto-encoder(CAAE algorithm).Analyze and learn from the excellent structure of the VGGNet model to provide the basis for improving the CAAE algorithm.A face natural evolution algorithm(ie,CAAE+ algorithm)based on semi-supervised adversarial auto-encoder is proposed.A method for evaluating the natural evolution model of human face is proposed.The natural evolution of human face is a double-constraint problem,identity constraints and age constraints.In this paper,subjective and objective experiments are conducted on the accuracy of age information and the retention of identity information.The experimental results show that the face images of other ages generated by the face evolution model based on the CAAE+ algorithm are more similar to the real face image in terms of identityinformation retention and age information accuracy,which proves that the proposed CAAE+ algorithm is more effective.