A new approach for shadow detection and compensation in color face images using within-class variance and effect evaluation

Nowadays, face recognition systems make significant contributions to human modern life. But, under some specific cases such as deep and soft shadows, the system performance will be degraded and the result is no longer correct. So, in this paper, we propose a robust and highly effective approach to detect all shadow regions from a face image and to make the compensation without yielding any visual artifacts in a face image. In order to detect all shadows, we make the within-class variance relationship between the background (skin) and foreground (shadow) information and find the optimum point for shadow-skin separation. For shadow compensation, many effect evaluations are performed based some shadow characteristics and then they are used as input parameters for a compensation function to reduce the shadow effects. The experimental results on indoor and outdoor face images demonstrate that our algorithm can work robustly and accurately under different lighting variations.