Research On Texture Description Algorithm For Iris Recognition

With the rapid development of information technology,biometric identification technology provides more effective means of protection for information security.As one of the biometric features,iris has the advantages of high complexity,uniqueness,stability and non-invasiveness,which makes iris recognition technology widely used in judicial,financial,security and other applications with higher security requirements.A complete iris recognition system usually includes iris image acquisition,iris image preprocessing,feature extraction and feature matching.Feature extraction is the realization of iris texture description,also known as texture characterization,which plays a key role in the accuracy and robustness of iris recognition.Therefore,this paper focuses on the description of iris texture,especially the stable characterization of iris texture.The main research contents are as follows:(1)In order to improve the characterization effect of iris texture and obtain stable expression of local texture information,an iris recognition algorithm based on multi-direction center-symmetric local binary pattern(MDCS-LBP)is proposed.Firstly,the correlation between pixel points is considered,and weighted gray values were calculated based on direction neighborhood and center neighborhood to provide more information for constructing feature mapping.Secondly,on the basis of center symmetry encoding,taking into account the role of center weighted gray value in iris texture characterization,the encoding method is improved to generate iris texture characterization information.Finally,the threshold binarization method is used to reduce the dimension of the features to obtain the binary feature template,and Hamming distance is used to calculate the matching scores between the binary feature templates to realize iris classification.Aiming at the deficiency of correct recognition rate(CRR)of evaluation index,a modified correct recognition rate(MCRR)is proposed.On the iris datasets of CASIA.V1,CASIA.V3-Interval,JUL6.0 and CASIA.V4-Lamp,the MCRR of the proposed algorithm is 99.86%,99.94%,99.64% and97.97%,respectively.Compared with the Daugman’s algorithm,equal error rate(EER)is reduced by 0.97%,0.58%,2.12% and 5.93%,respectively.Compared with SCCS-LBP operator,the equal error rate is reduced by 0.22%,0.03% and 0.52%,respectively.The experimental results show that the proposed algorithm can effectively characterize the iris texture information,and has good recognition performance and robustness.(2)The decline of iris image quality and elastic deformation of iris texture decrease recognition accuracy and algorithm robustness,which is caused by the inadequacy of the characterization of iris texture information by a single iris texture characterization model,especially as intra-class and interclass distances are too close or overlap,resulting in the failure of correct classification.To solve this problem,an iris secondary matching algorithm based on MDCS-LBP operator and 2D-Haar wavelet is proposed.Firstly,MDCS-LBP operator is used to obtain iris texture characterization in spatial domain to complete the first matching.Secondly,for iris images whose intra-class distance and inter-class distance are too close or overlap in the first matching,2D-Haar wavelet is used to obtain iris texture characterization in frequency domain for secondary matching,so as to realize the classification of iris images which are prone to cause errors in the first recognition.On the iris datasets of CASIA.V1,CASIA.V3-Interval and CASIA.V4-Lamp,the MCRR of the proposed algorithm is 99.92%,99.96% and 98.77%,respectively.Compared with 2D-Haar wavelet algorithm,the equal error rate is reduced by 1.13%,0.38% and 1.81%,respectively.Compared with MDCS-LBP operator,the equal error rate is reduced by 0.02%,0.04% and1.14%,respectively.The experimental results show that the proposed algorithm has high accuracy for high quality iris image recognition.It can effectively suppress the decline of recognition accuracy and improve the robustness of the algorithm in view of the decline of iris image quality and elastic deformation of iris texture.