Multi-view face detection and pose discrimination in video

Symmetrical frame difference was used to detect the motion area and a skin-color segmentation algorithm based on BP neural network was used to extract the face candidates. Then, a combination of multiple neural networks, each corresponding to a certain range of face orientation, is used for face verification as well as for pose discrimination. Experimental results show that the algorithm is robust for human faces under deep-view rotation, and is adaptive to various lighting conditions and different face sizes.