Face Recognition Based On LBP And Stacked Autoencoders

Face recognition is one of the most active topics in pattern recognition.How to effectively extract the feature of face image in face recognition process is the focus of the work.The traditional machine learning technology has great limitations in dealing with the natural data.The construction of the machine learning system requires the manual design of the feature extractor,and the deep learning method is automatically learned by the machine.Stacked autoencoders is one of the deep learning algorithms,built by a number of simple non-linear modules,with more powerful image feature extraction and classification recognition function.However,SAE ignores the local structure features of faces when it extracts face feature in face recognition.In this paper,a face recognition algorithm based on LBP and SAE was proposed in this paper.The experimental results on Extended Yale B face databases demonstrate that the proposed method is robust to illumination.And it has a better recognition performance compared to traditional algorithms and standard SAE.And the weight is added to the local binary mode algorithm.The main works of this paper are as follow:(1)Local Binary Pattern is an effective texture description algorithm with the advantages of rotation invariance and gray invariance.In this paper,we proposed a face recognition method based on weighted LBP.The weights are added by two methods.One is based on a priori knowledge.One is based on entropy-weighted.Experiments show that both methods can achieve higher recognition rate,but the advantages of the second method is more obvious.(2)To solve the problem that SAE ignores the local structure features of faces when it extracts face feature in face recognition,a face recognition algorithm based on adaptive LBP and SAE was proposed in this paper.The adaptive LBP algorithm has a strong robustness to the illumination and attitude changes.This algorithm adds the adaptive weights to the LBP algorithm first,and combines adaptive LBP and SAE.Experiments show that the proposed algorithm can enhance the minutiae of face images,and it is very robust to the illumination and obtain a high recognition rate.