컬러 모델을 이용한 얼굴영역 및 동공검출

Abstract

There have been many interests about real-time face detections in the pattern recognition. The face detection based on skin color model has been actively researched to detect the face quickly. But most of the skin color models are the single channel models to eliminate the light source or the adaptive models to reduce the features of color channels. The goal of this paper is to propose a multi-channel skin color model, in which Hue, Cb, Cg colors are used to remove light source and Red, Blue, Green colors to detect the face color effectively. For this purpose, we develop a robust pupil detector to facilitate the detection using active illumination which take account of the retro-reflectivity property of eyes. The location of the pupil candidates are found and computed. Then the candidates are filtered and grouped into pairs that correspond to face region with heuristic rules. For robustness of the face detection technique, we develop a dual mode face tracker to initialize with the most salient face region. Also, a region restricting technique is used to detect eyes region and then the gray-level threshold algorithm in the detected region to confirm it.