A simple and real-time eyes detection without training procedure

A fast algorithm for eyes detection based on the symmetric and geometric relationships of human-eyes region in gray-level images is presented. In order to reduce the search effort and eliminate the noise impact, we first perform the edge processing with Sobel filter. We then make use of the characteristics of symmetric relationship of human-eyes region to find out the possible regions of human eyes. Moreover, we use the geometric characteristic of human-eyes region to preliminarily exclude the region without eyes and reach the locations of possible eyes regions as quickly as we can,. Finally, to merge all possible regions with eyes into one and to exclude the regions without human eyes effectively, the eyes verification process is also proceeded. The algorithm is fast and can be used in real-time application. In Static image experiment. The correct rate is 96.4%(2086/2164) for FERET database. It takes about 0.015 second for detecting each image. The proposed algorithm Without Training Procedure has been tested on various real images and its performance is found to be quite satisfactory. The demonstration of our work is available at: http://www.ee.tku.edu.tw/~dsp/tkufd.