Research On Iris Localization And Recognition Algorithm

Iris recognition is an identity recognition technology which use one’s iris region in the eyes.Since iris is unique, high stability, high security, and there is no need to contact or offend thecandidate in iris recognition system, over the past few years, iris recognition has beenextensively studied and supported. However, due to the application environment of irisrecognition is very complex, it is difficult to locate iris in a variety of factors interferenceenvironment. Also, there are some problem while combining iris recognition technology totemplate security. Because the iris feature generated from existing iris recognition technologyis not rotation invariant, usually iris recognition system has to store a plurality of templates,which is not suitable for combination with existing iris template security framework. In thepaper, we focus on these two shortage parts in iris recognition, and proposed several improvedmethod. And the main work of this paper is mainly reflected in the following aspects:Firstly, we presents two iris localization algorithm for the ideal (constrained environment)and non-ideal (non-constrained environment) iris data. In constrained environment, we use theidea of region growing to locate the pupil coarsely, followed by the use of Canny edge operatorand Least Square method to fitting the pupil edge to narrow the parameter space of Houghtransform for accelerating pupil localization, finally we use the Intergrodifferential operator tofinish the iris localization. And in non-constrained environment, we use a gray level growingway, then use the global minimum eccentricity to stop the iterative to get a complete pupilregion. Also we use maximum differential point set to complete the iris localization in the caseof multiple interference in human eye image. And the proposed iris localization algorithm forconstrained environments accurately locate99.34%iris on CASIA-V3-Interval database, whilethe iris localization algorithm for unconstrained environment accurately locate99.55%iris onCASIA-V4-Thousand database.Secondly, in order to avoid the cumbersome steps in traditional iris recognition algorithms,which is multi-angle normalization and shifting matching for feature comparison, and to makethe iris feature extraction suitable for iris template security framework, in this paper we proposean iris feature extraction method based on PCET moments to generate the iris feature which isrotation invariant. This method use particle swarm algorithm to optimize multiple two- dimensional Gabor filter. Then, we filter the iris normalization image under this set of two-dimensional Gabor filter. And in the filtered results PCET moments are calculated, then featureselection will be performed on these PCET moments to generate the iris template. Finally, weuse the iris template for identification and calculate FRR, FAR and EER for our iris recognitionmethod. Experimental results show that the proposed iris feature extraction algorithm,compared to similar algorithms with rotation invariant，has better recognition results on irisdatabase and the equal error rate is1.19%.