An Accurate Eye Detection Method Using Elliptical Separability Filter and Combined Features

This paper presents an accurate eye detection algorithm using elliptical separability filter and combined features of eyes. First, a histogram back-projection method is utilized to extract the rough face region, and then iris candidates are detected by using elliptical separability filter developed based on Fukui et al.’s separability filter. By calculating the similarities of pairs of iris candidates, we determine the pair of iris, which has the largest similarity among others. The similarity of a pair of iris candidates consists of the separability of the pair of iris candidates, the similarity between VQ histograms and normalized correlation coefficient between the region including the pair of iris candidates and eye template. Experimental results show the iris detection rate of the proposed algorithm of 95.2% for 516 images of 86 persons without spectacles in the AR database.