Research On Iris Recognition Algorithm Based On CS-LBP And Adaptive Neural Network

With the development of modern science and technology,some drawbacks appears in traditional authentication.Biometric recognition represented by iris recognition has become a new research field and is gradually accepted.It opens up new ideas for the identification of people.Iris recognition is mainly through the extraction of iris texture information,matching an iris with a pre-stored iris template and identifying the identity of current person.Comparing with face,fingerprint,gait and other biological features,the iris feature is stable and doesn't change with age,so the accuracy of iris recognition is high.It is very suitable for use in some high-secret places or some special situations such as banks,prisons,medical insurance and so on.Iris recognition is divided into one-to-one recognition and one-to-many recognition.Rather than identifying one person from many candidates,this paper focus on one-on-one recognition comparing the iris to be recognized with only one stored iris.With feature extraction,spatial domain algorithms represented by Local Binary Patterns(LBP)are widely used.Compared with other frequency domain recognition algorithms,this way of discribing the relationship between gray levels can better reflect the changing trend of iris texture,and is more intuitive.Neural networks are often used in iris recognition,the optimization of neural network is an important part of neural network design.In order to determine whether two irises are the same or not,the Center Symmetric Local Binary Pattern(CS-LBP)algorithm is used in this paper to extract the variation of iris gray levels as features,and the BP neural network is used to recognition combined with blocking Hamming distance.The connection weights of BP neural network are adaptively optimized by chaos operator-selection&mutation-particle swarm optimization(C-SM-PSO)algorithm to improve the global search ability,to jump out of local optimum,and to improve the accuracy of iris recognition.Rather than changing the structure of neural network,this algorithm can adjusts connection weights of neural network when applied to different iris libraries,and has good versatility.CASIA-V1 and CASIA-Iris-Twin Iris Library of Chinese Academy of Sciences are used to evaluate the performance of C-SM-PSO algorithm.CASIA-V2 and CASIA-Iris-Lamp Iris Library of Chinese Academy of Sciences are used to compare this algorithm with some classical algorithms.Evaluation indicators are the correct recognition rate(CRR),the equal error rate(EER),and the ROC curve.Experimental results show that the proposed algorithm improves the recognition rate with lower equal error rate.The ROC curve is closer to the horizontal and vertical coordinate axes showing stability,and the structural versatility has been greatly improved.