Abstract
Face detection in a natural scene is an essential step in automated human face recognition. The eyes-strip of a human face plays the most important part in the detection process, since this area is not affected by hairstyle. Geometric techniques for feature extraction have the problems of sensitivity to lighting conditions and facial expressions. The weakness of the geometric techniques could be easily avoided using neural networks. In this paper, a learning vector quantization (LVQ) neural network is trained to detect the eyes-strip in skin-like areas only. First, a modified color system is used to extract skin-like areas from a natural scene. The fuzzy C-means algorithm is then used for thresholding the hue weighted chroma image. The skin-like area builds the basis of further search for human facial features. A LVQ network is trained on different eyes-strips and other areas of both facial and nonfacial images.