Development of a geometrical algorithm for eye detection in color images

Abstract

The applications of eye detection have grown significantly during the past decade. The objective of this paper was to develop a new technique for eye detection in color facial images. The implementation of this technique was composed of three steps. First step is using skin color information to detect a face in color images. The face region would be segmented from the background. Next step, illumination based method (chrominance components and luma component) was employed to find the possible location of the eyes. Finally, to locate the exact position of the eyes, the flexible thresholding method and new eye geometrical tests were utilized. The proposed technique was tested using Aberdeen facial images from Psychological Image Collection at Stirling (PICS image database). The simulation results indicate the detection rate of 93.2% for frontal facial images and 85.7% for rotational facial images.

An efficient hybrid eye detection method

“…Phromsuthirak et al [15] 473 frontal facial images from PICS 93.2% 14 rotational facial images from PICS 85.7% Feng and Yuen [2] MIT AI laboratory face database 92.5% Wu and Trivedi [8] 317 images from FERET 92.43%…”

“…The method had high detection rates of 96.98%, 94.29%, 98.65%, and 95.32% on the FERET, Aberdeen, IMM, and CVL databases, respectively, and robustness to different poses, light conditions, and glasses. Phromsuthirak and Umchid [15] first adopted the EyeMap algorithm to locate possible eye regions, and then the correct eye regions were determined using the proposed geometrical test methods. The evaluation results showed a detection rate of 92.3% for frontal facial images and 85.7% for rotational facial images.…”