The Application Of Local Binary Pattern And Convolutional Neural Network In Multi-pose Face Recognition

Face recognition is to judge whether two people are the same person or identify a specific person from the face databases.That is,1:1 verification and 1:N verification,rather than just detecting a face.Computer vision and digital image processing are all searching for the core technology behind face recognition.This has attracted the attention of peers in the worldwide.Face images are grayscale,frontal,half-occluded,different lighting and etc.This article researches on multi-pose face meeting in real life and has a referenced influence on really achieving face recognition.The main work of this article is to solve effectively image features extractions and improve the multi-pose face recognition rate.At present,there are many multi-pose face feature extraction algorithms,which are mainly divided into traditional and modern methods.The commonly used traditional face recognition method is the local binary pattern(LBP),however the traditional face extraction feature by LBP is poor,and can not achieve the desired effect.The modern face feature extraction method takes the deep learning technology as the core.And deep learning technology is based on a convolutional neural network.Through its learning of facial images,many structural models are published by researchers.However,most networks are both too complex and face recognition rates are low.These problems greatly affect human faces recognition of development.In view of the above problems,this paper presents two multi-pose face recognition algorithms,and the specific contents are as follows:(1)A multi-pose face recognition algorithm combined with discrete wavelet transform(DWT)and local binary pattern(LBP)was proposed.Issue on extracting features by LBP is too single,so this paper makes DWT decompose the input face images to get low-frequency sub-band and high-frequency sub-band,and the LBP extracts for each sub-band.The experimental results show that DWT+LBP multi-pose face recognition algorithm is superior to PCA and LBP,and this idea in this paper can be broadened to study other related fields.(2)A multi-pose face recognition method based on multi-scale convolutional neural network is proposed.For most of the current convolutional neural networks are single-layer neural networks,and the ability of face feature extraction is weak.It isproposed in this paper that the input image is simultaneously subjected to a two-layer convolutional neural network feature extraction.Experimental results show that the algorithm of multi-scale convolutional neural network face recognition is superior to other single structure methods.This point of view provides ideas for changing the network model.The traditional face recognition method is suitable for a relatively small number of datasets.This paper carries out DWT and LBP's algorithm to train and test on ORL face datasets,and the recognition rate reaches 97%,which is higher than other traditional face recognition methods.Deep learning face recognition method is suitable for a large number of datasets,and this paper conducts multi-scale convolutional neural network to train on CASIA-Web Face and test on LFW.Experimental results show that the recongnition rate of multi-scale convolutional neural network reaches 96.3%,which is higher than other network structures.Through the above experiments,the effectiveness of this research method is proved.