ROTATION INVARIANT EYE DETECTION

摘要

In this paper we propose a new algorithm to detect eyes in color images. The algorithm first tries to find eyes from hole-pairs in the face region. But, if none of hole-pairs fit eyes, the algorithm selects eyes from valley segment-pairs. Valleys mean darker regions than their surroundings. In the proposed algorithm, valleys are divided into segments so that each segment corresponds to a facial feature or a part of a facial feature. And, the algorithm computes symmetric measures of lines using the valleys in the face region and uses them to pick up eyes. By experiments, we verified that the algorithm detects eyes with high success rate even from faces whose image-plane rotations are unknown.