Research And Implementation Of Face Recognition Algorithm Based On LBP And HOG

Face recognition technology is a kind of biological recognition technologies which use the person's facial features information for identity authentication.With the advantages of non-contact,reliable and intuitive,face recognition technology has been extensively applied in face recognition attendance system,criminal detective,b,customs and border control etc.The process of face recognition mainly includes image acquisition,image preprocessing,feature extraction,feature matching and so on.Among them,feature extraction is the most critical stage of face recognition.This paper focuses on the feature extraction of facial feature.The Local Binary Pattern and Histogram of Oriented Gradient can effectively describe the local facial feature.LBP can effectively describe the local texture features of the image and its computational complexity is low.Besides,it has a better robustness to the change of the light and expression.HOG can significantly describe the edge and shape information of the image.They have mutual complement in describing facial features.In this paper,the fusion algorithm of LBP and HOG is studied and improved.The main work and innovation are as follows:(1)The features of LBP and HOG are studied,and a face recognition weighted fusion algorithm based on them is designed and implemented.The robustness of the fusion feature is enhanced by assigning the appropriate weighting coefficients to the LBP feature and the HOG feature.Experiments show that the algorithm can accelerate the recognition speed as well as improve the recognition rate.When the image is feature extracted in the case of fewer blocks,the recognition rate can be effectively improved by changing the weighting coefficients.The optimal recognition rates of FERET,ORL and GT face database were 97.33%,94.58% and 79.43% respectively.(2)The feature extraction process of multi-level histogram sequence is studied,and on the basis of which,a new algorithm based on the combination of fusion feature of multi-level histogram sequence CS-LBP(MHSCSLBP)with HOG and Fisherface is proposed.The Fisherface algorithm is chosen to reduce the fusion feature dimension,not only can it remove a lot of redundant information,but also it preserves and enhances the useful information of the feature.The proposed algorithm has better stability than single local feature,and can still get higher recognition rate under lower fusion series.At the same time,with the combination of Fisherface for dimensionality reduction,the algorithm has the advantage of controllable feature dimension.Under the FERET,ORL and GT face database,the optimal recognition rates were 96%,97.92%,and 87.43%,respectively.(3)Based on Android NDK development mode and Open Source Computer Vision Library,the face recognition system based on Android platform is achieved.And functional testing and performance testing are conducted on the system.The system can be used for real-time face detection and verification.