Single Sample Face Recognition Research Based On Monogenic Filtering And LBP

With the development of science and technology,face recognition has become a popular biometric technology in the field of computer technology research.It has the unique advantages of non-contact and high confidentiality compared with other biometrics.In recent years through the tireless efforts of people at home and abroad,face recognition method recognition rate is also rising.But in some practical scenes,such as personal ID card,public security handling case or customs inspection,only a single face to face comparison,this time the classic traditional face recognition method recognition rate is not high,and even some will quickly decline.Therefore,single face as a training sample of single sample face recognition research direction has a very wide prospect.A complete face recognition system process is: first detect and collect face images,preprocess the face image,and then extract the features of the face image after processing,and finally match and identify;identify the process of each Links are essential to play an important role,but relatively speaking,the decisive role is the characteristics of the extraction is good or bad.The local feature of monoling analytic information extraction with rotation invariance not only contains more identification information,but also has the advantages of low computational complexity.Localized binary model(LBP)is a simple texture description operator used to describe texture,but the description of facial features is not sufficient.This paper will study the feature extraction based on single-shot filter under single sample condition.The main tasks include:(1)A single sample face recognition method based on the same amplitude and phase of monogenic signal is proposed.In this algorithm,the monogenic amplitude,monogenic direction and monogenic phase image of different scales are obtained,and performed same direction amplitude feature detection and phase quantization to get the block histogram.Then,the each subblock of histogram weighted according to the local information entropy of each subblock.And then the weighting histogram of each sub-block is connected as the eigenvector of the final image and sent into the classifier to identify.Experiments result on YALE face database and ORL database show the proposed algorithm is effective.(2)A new method of face recognition based on Monogenic dominant orientations and center-symmetric local binary pattern(MDOCSBP)is adopted to single sample face recognition.Firstly multi-scale monogenic filter was used to get monogenic local amplitude and direction information of a face image,and the amplitude and derection information were used to calculate dominant orientations to get Pattern map of Dominant Orientations;Then center-symmetric local binary pattern(CS-LBP)is proposed to encode Pattern map of Dominant Orientations to get MDOCSBP feature maps;finally,MDOCSBP feature maps at different scales are divided into several blocks,and the concatenated histogram features calculated over all blocks is used for the feature descriptor of face recognition,and the recognition is performed by usingthe histogram crossr.Experimental.Experimental result on AR face database and Extend Yale B face database show the proposed algorithm is effective.