Skin Colour Based Face Detection

Abstract. This paper describes a new approach to face detection. A colour input image is first processed using neural-networks to detect skin regions in the image. Each neural network separates skin and non-skin pixels on the basis of chrominance information. The skin-colour classifier employs the committee machine technique, which improves skin colour detection by combining the classification results of a set of multilayer perceptrons (MLPs). The skin colour classifier achieves a classification rate of 84 % compared to 81 % for the best individual MLP classifier. The output of the committee machine is processed by a 2D smoothing filter before being converted into a binary map using a threshold. Finally, several post-processing techniques based on shape and luminance features are proposed for rejecting non-facial regions. 1 Introduction wavelet packet analysis is employed in [5] to extract texture features stored in the intensity Face detection, which aims to detect the presence plane gr) of the image, which are then classified and subsequently the position of a face or faces in into face or non-face using the Bhattacharrya an image, is an important task in automated facial distance. image analysis. The result of face detectio