DEVELOPMENT OF A SEGMENTATION ALGORITHM FOR IRIS RECOGNITION BASED ON ELLIPTICAL APPROACH

Abstract

Iris, unlike other biometric trait possesses unique features, which are highly randomised. It consists of various pigmented structures like crypts, collarrette, ridges, minutiae arranged in compact manner, which do not change throughout lifetime. This work appraised the research effort of Daugman, who pioneered iris automation system as means of human recognition. In Daugman’s work, the region of interest (iris) in eye images is considered to be bounded by papillary boundary and sclera/iris boundary. The two boundaries were considered to be circular geometrical structures because of the ease in computation and employed integro differential operator as a suitable circular finder to detect the two boundaries on any given eye image contour. The regular and involuntary dilation of pupil stimulated a better description of the papillary and sclera/iris boundaries to be elliptical and not circular description held earlier by Daugman. This new position consequently resulted to the development of a segmentation algorithm for iris recognition based on elliptical approach (Ellipto integro differential operator). This consequently leads to a new boundary finder in iris recognition system. The two algorithms were implemented at 4,8 and 16 operational octave levels. Performance evaluation indices utilised in this work include Dev  (deviation) and Amplitude, b b Pupil and Iris and p i Max and Max derived from experimental results. The implementation results are organised into graphs and tables. The integro differential operator (IDO) serves as contour fitting for the circle while the ellipto differential operator (EIDO) serves as ellipse boundary finder. From the results, the average maximum contour integral on the pupillary and sclera/iris in the eye obtained in the eye image domain is 0.017370913, 0.0116584 for IDO, 0.011011830, 0.003090348 for EIDO at 4-octave level while 0.00102954154, 0.028498258 for IDO, 0.00101383754, 0.0013648786 for EIDO at 8-octave level and at 16-octave level the average posted are 0.0013322312, 0.004538600 for IDO and 0.0013185924, 0.004485164 for EIDO. The average value of the amplitude computed from experimental results in appendices I and II gave 35.4 and 33.6 to IDO algorithm and the proposed EIDO respectively and deviation and Dev  at 4 octave for IDO stood at 0.04706 and 0.04536 for EIDO. The overall performance at sixteen operational octave level reveals almost 50-50 performance at both EIDO and the Daugman’s IDO algorithm. From the foregoing, the improved algorithm produced a better performance than IDO and consequently could therefore be considered as a better alternative contour fitting method compared to the existing approach.