Palmprint Recognition Approach Based On Complex Network And Muilti-wavelet Features

Nowadays,more and more attention has been paid to the problem of information security.Compared with other biometric features,palmprint recognition technology is more simple in sampling,and it's easy to be accepted by users,what's more,it does not require high precision instrument in sampling and identification.In the past,all the information of palmprint image was used in palmprint recognition technology.we usually,in the fact,the main factors determining the palmprint identity information are the line and wrinkle features,which are used in these paper.In extracting the line and wrinkle features of the palmprint,the result will be affected seriously because of the image noise.CL multi-wavelet has excellent characteristics,which not only can effectively remove high-frequency noise,but also extract the line and wrinkle features in different directions.So,the palmprint image is transformed into binary valued texture image by CL multi-wavelet and the mean window method,which realize the elementary palmprint features extraction.In addition the complex network can describe the topological structure of the extracted binary image.a series of dynamic evolution of networks are generated by choosing different thresholds.Thus the description of the palmprint texture can be realized through the description of the network.Two palmprint recognition approaches based on multi-wavelet and complex network are proposed in this paper.The main research works are as follows:1)A elementary palmprint image feature extraction method based on the CL multi-wavelet and the mean window method are proposed.The low,horizontal and vertical frequency components?LL1,LL2 and LL3 sub-images?corresponding to the prefilters in the CL low frequency component are extracted first.And then the local binary texture features are obtained using the mean window method,which can keep the palmprint line and wrinkle information well.2)The basic concepts of complex network are introduced and the metrics are selected to describe characteristics of the network.3)A palmprint recognition method based on CL multi-wavelet and complex network are proposed.The binary texture images?BLL₁,BLL₂ and BLL3 sub-images?are obtained by the CL multi-wavelet decomposition and mean window method first.And then,the extracted binary texture images are combined to form the palmprint elementary features.Third,a series of dynamic evolution of networks are generated by choosing different thresholds,whose statistical characteristics of the networks?the average degree,the biggest degree,the standard deviation and average energy of degree?are computed as the secondary features.Fourth,The final features are obtained using the LDA method.Finally,the nearest neighbor classifier is used to classify.4)An improved multi-wavelet feature extraction method based on multi-wavelet and local optimized complex networks are proposed.In the method of 3),three sub-images are used in the network modeling at the same time,and the information of their own structures and relative positions are also included to some extent.By observing the BLL₁,BLL₂ and BLL3 sub-images,It can be seen that the sub-image BLL₁ can represent the line and wrinkle features of palmprint better than BLL₂ and BLL3.So the sub-image BLL₁ are analyzed further.The local dynamic evolution complex networks are generated for the BLL₁ sub-image and its four sub-blocks respectively.By means of analysis and experiment,the average and standard deviation of degree are determined as the network metrics.In this paper,CASIA?of Automation of Chinese,Academy of Sciences Institute?library for the experiment,through the comparison with the traditional method proves that two proposed method can effectively perform palmprint recognition.Especially,the improved method in 4)can reduces the feature dimension and significantly improves the recognition performance than the proposed approach in 3).In this paper,CASIA?of Automation of Chinese,Academy of Sciences Institute?library is used for the experiment.The experiments show that two proposed palmprint recognition methods are more effective than the traditional methods.