Eye Detection Method for Face Images Using Circular Filters and Eye-Weight Mask

초록

This paper introduces a method for detecting eyes in face images. Firstly, the detection of face region is carried out by using the skin-color cue to yield a skin and face masks. From the face mask, the grayscale face image is extracted and adaptively thresholded by considering the distribution of the facial skin. Next, the circular filer, which is specifically designed to detect circular shapes in images, is applied to this binary image. An eye-weight mask generated from the binary face image, which puts small weights on the near-boundary regions and eyebrows as well as beards if present, is weighted to the filtered image to produce an eye-peaked image. Finally, the eyes are located by searching for the peaks in the resulted image. Experimental results conducted on a face database including variety of head poses show that the proposed method is robust to different head poses.