Low complexity Iris recognition using contourlet and curvelet transforms

Abstract

Biometrics refers to automatic recognition of individuals based on their physiological and behavioral characteristics. The human iris has mesh-like textures with numerous overlays and patterns. It provides a very high degree of uniqueness, of about 1 in 1072, determined during anatomical development. In addition, it is visible, thus making it quite attractive for being used as biometric features for noninvasive person identification systems. Various methods have been proposed in the literature for efficient iris recognition. The most promising results are reported by those using iris information extracted in the timefrequency domains. However, many of these methods show high accuracy at the expense of being computationally expensive. The high computational complexity results from (i) a considerable amount of time being consumed during feature extraction, mainly spent for the preprocessing of an iris image and (ii) the use of iriscodes of long length that increases the time required for matching. Considering the huge volume of iris data to be compared in practice, reducing the computational complexity is important. In recent times, the contourlet and curvelet transforms have emerged as highly effective time-frequency representations of natural images. However, few methods are available in the literatures that exploit these transforms for iris recognition, albeit reporting good performances for a rather limited number of subjects. In this thesis, a low complexity technique is proposed for iris recognition in the contourlet as well as curvelet transform domains. The proposed method does not require the detection of outer boundary and decreases unwanted artefacts such as the eyelid and eyelash. Thus, the time required for preprocessing of an iris image is significantly reduced. The zero-crossings of the transform coefficients are used as iris codes. Since only the coefficients from approximation subbands are used, it reduces the length of the code. The iriscodes are classified using different classifiers that include the correlation coefficient, hamming distance and K-nearest neighbor (KNN) classifier. Extensive experiments are carried out using a number of standard databases such as CASIA-V3, UBIRIS.v1 and UPOL. The iris images used are obtained under different lighting conditions and time interval. The results reveal that the proposed method using the curvelet transform provides a very high degree of accuracy (about 100%) over a wide range of images with a low equal error rate (EER) with a significant reduction in the computational time, as compared to those of state-of-the-art techniques.