Research On Face Recognition Method In Complex Environment

Face recognition has been applied widely to some specific business scenarios,such as access control,video surveillance,etc.However,face recognition is affected by many factors,for example,in unconstrained face recognition,the face images may have many variations,such as low resolution,pose variation,complex illumination and motion blur.In this paper,we present a pose-robust feature learning for pose-invariant face recognition and a novel framework of dependency model is proposed upon the assumptions that different feature extraction methods are dependent to each other:1.Pose variation is the important factor that affects the performance of face recognition.Almost all the methods need to calculate the deflection angle by manual or deep learning to handle pose-invariant face recognition(PIFR).We propose a new network including a stem Convolutional Neural Network(CNN)and a mapping module to automatically project different poses into the deep features under the shared latent subspace.The stem CNN and the mapping module could be trained together in an end-to-end manner.And by using the mapping module,the calculation of deflection angle is not necessary.According to the experiments on two databases,the superiority of the proposed network over the nine state-of-the-arts is demonstrated on face identification task.2.Almost all feature fusion methods run into two problems,feature dependency and fusion assumption on the posterior distribution.To solve the above problems,we propose a new method for multiple feature classifiers face recognition,the dependency between multiple feature classifiers was modeled by Hidden Markov Mode(HMM)in non-cooperative and uncontrolled environment.According to the experimental results on five databases,the proposal shows promising performance under unconstrained environment,especially in artificial and multiple noisy interferences.