Research On Face Image Aging Method Based On Recurrent Neural Netword

With the age increasing,the geometric features of the human face will gradually change.This process is an irreversible natural law.It is mainly affected by factors such as genetic genes and living environment.Therefore,accurately aging face images plays an important role in the search for missing population,cross-age face recognition,and age estimation.This thesis studies the aging method of face images and evaluates the result of the aging method.In this thesis the face aging method proposed includes face feature points calibration,face poses correction and the face aging modeling based on recurrent neural network.Face features include facial contours and five features.Firstly,the supervised descent method algorithm is used to calibrate the feature points of the face,and the face pose is judged by the position of the feature point.Thereby laying the foundation for the face pose correction;then the face pose correction is performed.Face pose affects face features,so face pose correction is necessary.The face rotation angle formula is obtained by combining the facial feature points with the linear regression equation,the face rotation angle is determined by the position of the face feature points,then the face is corrected by affine transformation;Finally,adopt the recurrent neural network to bulid the Face aging model.Face features are obtained by performing singular value decomposition algorithm,and these feature values are used reconstruct the face images,these images input into a recurrent neural network for aging modeling,thereby obtaining an aging face image.In this thesis,the structure of the recurrent neural network is divided into two layers.The underlying network encodes face features.The top-level network decodes the underlying output into an aging face image.Through training,the network parameter matrix weights and offsets are obtained.Then the face image is input to the trained network to get an aging image.In order to evaluate the aging effect,the deep convolutional network will be used to recognize face.The deep convolutional neural network is trained by the labeled experimental samples.The aging face image input into convolutional neural network to be recognized,the recognition rate is 88%.It proves that the RNN-based face aging method works well.