Face Recognition Based On Bit-plane Decomposition

Research on face recognition began in 1960.With the continuous development of computer and image research,it began to increase after 1980.After 1990,the United States,Germany,and other countries had some preliminary applications.Whether the face recognition technology can be successfully implemented depends on whether or not an excellent algorithm that can be practically applied is designed,and has a good recognition effect and a faster recognition speed.The face recognition field covers a variety of advanced technologies including machine learning,artificial intelligence,pattern recognition,and video image processing.It is an advanced biometric identification technology.The application of face recognition technology reflects the further development of artificial intelligence from weak to strong.In the case of authentication,law enforcement,etc.,the tested object can only provide one face image as a training sample for the algorithm.In the case of single sample,the recognition effect of the traditional face recognition method will be greatly reduced or even invalid.To solve the problem of poor recognition performance of the traditional face recognition algorithm under single sample environment,the bit plane of the face image and the local binary pattern are combined to expand the sample information and extract texture features better on this paper.The main contents of the thesis are:1.A single-sample face recognition algorithm combining improved CSLBP and bit-plane decomposition is proposed.Firstly,we use improved CSLBP to extract feature information of the face to obtain two texture images with different radii.Then,each texture feature image is decomposed into 4 bit plane images.Finally,8 feature images are fused in series,and the nearest neighbor classifier is used to perform the fusion.Classification identification.Experimental results on AR,CAS-PEAL,and Extend Yale B face databases show that the proposed algorithm has better recognition effect,faster recognition speed,and robustness to changes in illumination and expression.2.Two face recognition algorithms based on weighted fusion of bit-plane features are proposed.The algorithm based on Contribution Bit-plane Feature Weighted Fusion(CBFWF)firstly decomposes the face image and introduces the concept of contribution degree to set the weight of each bit plane.The weighted fusion of eight bit plane images constitutes a new sample image,which is used to perform feature extraction together with the original image,then uses the nearest neighbor classifier to classify and identify.The concept of entropy is proposed by the algorithm based on the entropy bit-plane feature weighted fusion(EBFWF).Firstly,after decomposing the bit-plane of the face image,the texture features are extracted by using the LBP operator.The information entropy of each feature map is calculated,eight feature maps are weighted and concatenated according to different entropy values,and finally the nearest neighbor classifier is used for identification.Experimental data on the face database of AR and CAS-PEAL shows that CBFWF algorithm and EBFWF algorithm can extract more characteristic information,obtain a good recognition effect,and have a certain degree of noise resistance.