Research On Aging Face Recognition In Identity Authentication

With the growth of age,the appearance of the face will have a significant change.Such as the deepening of wrinkles,face changes.This requires that face recognition not only solves the problem of noise,rotation and distortion in the image of the same individual in the same period,but also resolves the recognition of face images captured by the same person at different times and in different environments.In recent years,the problem of declining face recognition rate caused by age change has become an important research direction in the field of face recognition.In order to solve the problem that the face recognition is sensitive to the age change,this paper proposes a aging face recognition algorithm based on hierarchical representation from three aspects of face feature processing,face aging simulation and face recognition.Details about this research are as follows:(1)A new hierarchical representation of face image is proposed.This article divides the human face into three layers:the global layer,the local layer and the detail layer.AAM model is used to get face feature vectors of face global layer and local layer,and we use SVM to get the regression equation of shape and texture with age.For the detail layer,the shear wave transformation is used to extract the detail layers of different age groups.(2)In order to get a better simulation of face aging effect,the face reconstruction model is established hierarchically.For the aging simulation of the global layer and the local layer,the regression equation of shape and texture change with age is obtained by using the support vector machine(SVM)training according to the corresponding face layer characteristics,and the face deformation process of two levels is completed.For the simulation of the detail layer,we mainly use the template of aging feature to transplant and replace.Finally,we get the final aged simulated image based on gradient pyramid image fusion algorithm.(3)Aiming at the problem that the traditional convolutional neural network(CNN)has slow convergence rate and low efficiency when training large-scale image set data,this paper proposes an improved CNN algorithm for face recognition.Firstly,the convolution kernel in the hidden layer is trained by a noise-reduction automatic encoder to obtain the feature output map of each layer.After subsequent hidden layer convolution,pooling and other operations,we get the final output characteristics.Finally,the method based on sparse representation and the feature map are combined to complete the face recognition process.