Abstract
Face detection in unconstrained environments is a difficult task with many challenges, due to the variations in face features, poses, lighting conditions and backgrounds. This paper presents a method to detect faces from the segmented regions of skin and hair color. A face is defined as the skin color regions of a collection of skin and hair color regions that satisfies two sets of geometrical rules. Hair color is assumed to be black. By using a transductive approach for skin color segmentation and selecting pixels of low intensities adaptively among achromatic pixels for hair color segmentation, reliable segmentation of skin and hair color regions from arbitrary images taken in common office environments can be obtained. Experimental results show that the proposed method has the capability to detect multiple frontal and non-frontal faces of various poses and scales, provided proper segmentation of skin and hair color regions are available. For video input, further processing of temporal information can be used to improve the detection performance.
