Fine Circular Pupil Localization

In this paper a simple, fast, and accurate method is proposed to localize the pupil. After analyzing eye images and their histograms, it has been found that the pupil area takes about 10% of the eye image area and mainly it is the darkest part. A threshold value was computed from the histogram of the image, the computed threshold is used to convert the eye image to binary image, the seed fill algorithm was used as region growing method to segment the binary image and locate the pupil as the maximum segment, then fill it with black color to remove any specular spot reflection. A circle fitting algorithm is used to locate the best pupil circle. The conducted test on the proposed system using 10 eye images, downloaded from CASIA-IRISV3-Interval database, indicated that the pupil area average error ratio is 0.6%, the pupil center average position error is 0.5%, and the pupil radius average error ratio is 2.1%.