Face detection based on knowledge and independent component

In order to improve the speed and robustness of detecting human faces in grayscale scene image, an approach that combines independent component analysis (ICA) with knowledge-based is proposed for face detection. A new detection rule is presented by analyzing face predigest models usually used in painting face. Moreover, in order to further reduce the time of face detection, the hierarchical detection processes and extended projection are used in the process of rough face detection. In addition, adaptive threshold selection is realized by maximizing between-class variance. Experimental results show that the new algorithm is feasible and efficient.