Human Skin Color Modeling, Segmentation and Correction for Face Detection

Human skin segmentation in colored images is closely related to face detection and face recognition systems as a preliminary critical required step where it is necessary to search for the precise face location. An important challenge for any image segmentation approach is to accommodate varying illumination conditions and different ethnic groups. We present a general approach for modeling, segmentation, and color correction of human skin that suppress false negatives errors in face detection problem. The proposed approach shifted from mono-skin model to multi-skin models using HSV color space. We use the color information explicitly to segment skin-like regions by transforming the 3-D color space to 2-D without losing of color information. Four skin color clustering models were used, namely: standard-skin model, shadow-skin model, light-skin model, and rednessskin model. Our approach combines pixel-based segmentation and region-based segmentation approach using the four skin-map layers and iterative merge process. To improve the detection rate, adequate attention was paid to automatic skin color correction. We present experimental results of our implementation and demonstrate the feasibility of our approach to be general framework which does not depend on additional preconditions and assumptions.