The face localization and regional features extraction

A double threshold Sobel operator to detect the edge of the face image and a square projection algorithm to locate the face in the edge image is presented in the paper. A technique extracting regional features is studied while a method based on geometric features is adopted to differ southern and northern faces. The experimental results show that the double threshold Sobel operator can present a clear contour of the edge image, lose little information and have a good immunity of interference. Furthermore, the square projection algorithm can offer an accurate face location, which is robust to the effect of face expression and partial occlusion. The probability that the southern and northern faces are correctly distinguished greatly increases when using the face feature extraction method described in this paper.