Robust Face Recognition Based On FPH Feature Extraction Framework

Face recognition is one of the most active research fields in computer vision and pattern recognition.In recent years,the face recognition system has been successfully applied to the information security,identity authentication,video surveillance etc.But in reality,the face image is affected by illumination,pose,expression and occlusion.Many researchers focus on how to extract the face features which are robust to the above mentioned factors.The FPH framework method which consists of feature maps,pattern maps and histogram is widely used,and has achieved remarkable results in robust face recognition.So far,the problems of why the FPH framework adopt such a combination and why it is so robust in face recognition has not been researched in depth.In this paper,we conduct research on the robust principle of FPH feature extraction framework and its application in face recognition.The main contributions are as follows:(1)Make in-depth analysis on the robust principle of FPH framework.In view of the feature map generation and pattern map coding,we concluded the rules of generating feature maps are the locality of feature extraction,the redundancy of feature decomposition and deep filtering.The principle and function of pattern map coding on feature map are explored,and the matching rules among feature maps and pattern maps are analyzed.Experiments on AR and Extended Yale B face database show the importance of each part in FPH.(2)Based on the robust principle of FPH,we propose a new method called Local Spherical Normalization(LSN),and embed it into PCANet,greatly improving the redundancy of features,and also enhance the ability of PCANet to filter noise in feature maps.By experiments in the AR and UMB-DB database,we prove that our method can further improve the robustness of PCANet features.In the future work,we will consider new methods of generating feature maps and pattern maps into FPH framework to improve the performance of robust face recognition.