Pose invariant face detection

In this paper, we present a novel method for pose invariant face detection in color images. The novelty in our method of face detection arises from the integration of evidence from various independent sources such as color, frequency response and geometric shape information. Skin color is detected in the Y-Cb-Cr color space using a RPROP neural network. The shape information is derived from a novel ellipse area criterion and then it is formulated to compute a probabilistic score of the connected components that represent the faces in the images. The third source is based on Gabor 2D filters that are used to obtain frequency signatures of faces in the images. The final likelihood of a face is a combination of the individual probabilities of color, shape and Gabor response.