Contact Lens Detection for Iris Recognition using Multiresolution Analysis

Abstract

Two-channel orthogonal and biorthogonal filter banks (FBs) have been used con- newlinesiderably in many signal processing applications such as data compression, de-noising, newlinepattern recognition, water marking, etc. In certain applications, linear phase filters newlineare highly desirable. However, linear phase cannot be obtained by orthogonal filters newlineexcept in the case of Haar FBs. Therefore, biorthogonal FBs are selected over or- newlinethogonal FBs in these applications.Moreover, there is a nearby association between newlineperfect reconstruction filter banks (PRFBs) and wavelet bases. Wavelets can be built newlinefrom iterations of PRFBs provided later satisfy additional constraint of regularity. newlineDifferent design techniques are developed for these two types of FBs. Although, both newlinethese types have their respective advantages, the biorthogonal FBs are preferred in newlineimage processing applications because of its linear phase characteristics. Biortho- newlinegonal wavelets can have symmetry and are associated with PRFBs. The critically newlinesampled biorthogonal FBs are designed with some desirable properties such as fre- newlinequency selectivity, regularity, symmetry, near-orthogonality etc. The design of FBs newlinewith these properties is a challenging task and hence received an attention over many newlineyears. Several methods including spectral factorization [1], [2], polyphase and lattice newlineschemes [3], [4] and parameterization [5] techniques have been proposed in the lit- newlineerature, to design two-channel PRFBs. These techniques have also been utilized to newlineoptimize desired FB traits. However, they generally do not provide global solutions newlinedue to the nonlinear nature of the design problem. Some of nonlinear design problems newlinecan be solved using optimization techniques, and global solution can be achieved. In newlinethis thesis we have designed biorthogonal FB using different optimization techniques newlineconsidering different properties of FB as objective function and constraints. The per- newlineformance of the designed FBs is evaluated using the contact lens detection (CLD) newlineapplication in iris recognitio