New Features Extraction Method for People Recognition on the Basis of the Iris Pattern

Abstract

Biometric people recognition methods are increasingly popular, yet there is no biometric authentication standard used in everyday life. Despite a lot of work on biometric people recognition methods, especially those based on the iris pattern, which is the subject of the author's research, there is still room for designing a new, optimal method, e.g. one that would be simpler in computation, have a shorter iris signature and good distinctiveness. In the paper the author proposes some iris database analyses (e.g. spatial entropy and average image analyses) in order to find input images parameters helpful for designing an iris recognition method. Then, a new iris features extractor based on reverse biorthogonal wavelet rbio3.1 is proposed, which is simple in computation, has a shorter iris signature (340 bits) and quite good discriminative power (d'=6.3, EER=0,6%) in comparison with Daugman's method used as reference. For experiments the UBIRIS database of 2105 images of 241 persons was chosen.