Research On Multi-source Heterogeneouce Iris Recognition

Identity authentication is a popular and important research hotspot in the field of security and information security,which is closely related to people's lives.As a representation of human identity,iris has important application values in terms of stability,uniqueness,and security.At present,most of the iris images are collected and identified on the same acquisition device.This way has high requirements on the use scene and application device,and has certain limitations.Therefore,due to different acquisition device,the problem of low intra-class similarity of iris,that is,the problem of multi-source heterogeneous iris recognition has become a challenging research.This paper mainly research on multi-source heterogeneous iris recognition method to achieve the effective feature extraction and accurate recognition.The research mainly in the following three aspects;(1)Preprocessing for multi-source heterogeneous iris images.First,use the quality evaluation method of multi-index fusion to screen out qualified iris images;then use the secondary positioning method to locate the inner and outer boundaries of the iris respectively;finally,normalize and enhance the iris,and intercept the iris area of interest.(2)In order to improve the feature-by-feature point extraction and matching methods in iris feature extraction,the multi-source heterogeneous iris recognition algorithm based on locality preserving projection based on manifold learning is proposed to extract the iris local texture feature structure on a region-by-region basis.The local nearest neighbor structure is established for the feature points with a large similarity in the original feature space to form a manifold.The feature map is used to retain the nearest neighbor structure with a large similarity,while maintaining the relative position of the feature points.Experimental results show that on the JLU-MultiDev data set,the locality preserving projection retains the iris texture structure effectively,reduces the computational complexity,and has a high accuracy rate of multi-source heterogeneous iris recognition.(3)To further improve the feature extraction method from region-by-region,the multi-source heterogeneous iris recognition algorithm based on stacked CDBNs-DBN is proposed.Find the position of the effective feature points by the displacement of the hidden unit,extract the local texture feature structure through the convolution layer,and reduce the feature dimension through the pooling layer.At the same time,the network establishes the relationship between the local feature structures through the full connection between layers matching relationship.Finally,the multi-source heterogeneous iris images are classified and identified by DBN.The experimental results show that on the IIT Delhi database,the stacked CDBNs-DBN model has high recognition performance and strong robustness,and can complete multi-source heterogeneous iris recognition tasks.