Towards segmentation of non-ideal iris images using optimization based multilevel thresholding
Satish Rapaka, Rajesh Kumar Pullakura · 2018
Segmentation/localization of iris from an image is a vital process in an iris authentication system because it affects the authentication accuracy. Segmenting iris images captured under uncooperative conditions is an even more difficult task because of the noise artifacts like occlusions, overlapping intensities, and specular reflections. In this work, a pre-segmentation step using PSO and DS algorithm based multilevel Otsu thresholding has been proposed to improve the segmentation accuracy. The pre-segmentation process isolates the iris from unwanted regions of the image. The resultant images of a pre-segmentation step are then segmented by employing geodesic active contours (GAC) incorporated by a novel stopping function. The proposed method accuracy is evaluated by considering standard databases such as CASIA v3 Interval, UBIRISv1, and MMU1.