Robust Eye Detection from Facial Image based on Multi-cue Facial Information

Abstract

Eyes are one of the most important facial features and eyes detection is a crucial aspect in many useful applications. Although many eye detection methods have been developed in the last decade, a lot of problems still exist also. We presented a novel and efficient method to locate eyes from face images based on multi-cue facial information. Firstly, we made use of the skin information to extract the facial region. In order to obtain the correct facial region and avoid the effect of extracted binary mask with concavities or spikes at the sides in some cases, we proposed an algorithm called as filter close operation. Secondly, we transformed the processed face image into second derivate and then horizontally projected the derivate. A filter algorithm was developed to smooth the derivate based on the penalized least squares minimization and then the horizontal positions of eyes were extracted. Finally, we determined the vertical positions of eyes by vertical gradient projection function developed in the paper. We have used three different face databases to test the validity of our algorithms. The experimental results demonstrated that our developed algorithm can accurately locate eyes from face images. Comparative study with some existing eye detection algorithms has indicated the superior performance of the developed algorithm.