A heuristic approach for shadow and light regions fast detection in face images

Face detection and recognition have become more and more popular, especially in the era of hand-held devices. As a result, many algorithms have been developed to process face images. However, many of those also have problems with uneven illumination effects, because images have been captured under various lighting conditions. In this paper, we introduce a heuristic approach for shadow and light regions fast detection in face images. The results will be used as clues for other correction algorithms. Within the available samples of the face region, we use the K-means algorithm to cluster pixels into shadow, light and light-balanced regions. Since the heuristic K-means method may generate misclassified pixels, we use image processing techniques to enhance the clustered results. Experiments conducted on the Caltech face dataset show that our proposed approach can robustly, totally and quickly detect shadow and light regions in face images.