Image Resolution Related Behavior for Iris Identification Based on Color Features

Abstract

The most used approach for iris identification is based on analyzing its texture by methods introduced and refined by John Daugman. In order to obtain very good rates of correct identification, a relatively good iris image is required, i.e. the iris must have at least a radius of 70 pixels in a nearinfrared monochrome image. In latest years, we have developed a method of extracting color features from color images based on the LAB color system. The method constructs complex histograms over the LAB color space, made of bins corresponding to optimized rectangular parallelepipeds. We tested our method to see the minimum acceptable radius of an iris image that can still be correctly identified in most of the cases. To this end, we degraded four times the very good iris images in the UPOL database. The results obtained over the five sets of images, with radii varying from 250 pixels to only 16 pixels for each iris, show that our method can produce very good identification results at iris radii as low as 32 pixels.

Keywords

• iris identification
• iris radius
• texture
• LAB color features
• spatial information