ROBUST EYE FEATURE EXTRACTION FROM FACIAL COLOR IMAGE USING IRIS COLOR COMPENSATION

Abstract

This paper presents a novel method of robust eye feature extraction from facial color images by considering the variety of iris colors. Given an eye window containing a single eye, the proposed method assesses the iris color tone based on the difference images between the red and the green channels and the red and the blue channels. A weighted scaling compensation method is then proposed for increasing the separability and homogeneity of the iris region. The extraction of the eye features is performed by an unsupervised K-means clustering on the compensated feature spaces. The eye corners are detected after eyelid fitting using a least mean square cost function. Experiments on a collection of eye images extracted from the FERET face database show evidence of promising performance from color facial images with variation in illumination, pose, eye gazing direction, and race.