A robust shadow and light region detection using within-class variance in face images

Nowadays, computer vision has become increasingly important in real world systems for commercial, industrial, and military applications. And, a facial recognition system is one of such computer applications for automatically identifying or verifying a human face from a video frames by comparing selected facial features from the image and a facial database. Unfortunately, some recent algorithms have many problems in their accuracy due to some effects of illumination changes such as shadow or light. For that reason, we propose a robust shadow and light detection using within class variance which helps to detect all shadow and light regions in a face image. These detected regions will be the input of some recovery systems to obtain the illumination-invariant images. In this paper, we also have an overview of all shadow and light regions in a human face image and classified them into many different regions based on their characteristics. Results on various indoor and outdoor sequences under illumination variations show the success of our proposed approach.