A Nonparametric Skin Color Model for Face Detection from Color Images

Abstract

This paper presents a novel approach to extract face and facial feature points from color image automatically based on a nonparametric skin color model. Most of introduced skin color models for face detection have lack of robustness for varying lighting conditions and need extra work to reduce such problem. To resolve the limitation of current skin color model, we utilize the Hue-Tint chrominance model and represent the skin chrominance distribution as a linear equation. Thus, the facial color distribution is simply described as a combination of the maximum and minimum values of Hue and Tint components. The decision rule to detect skin region is simplified by measuring the distance between the skin chrominance distribution function and measured input chrominance. In order to extract facial feature points defined by MPEG-4, the minimal facial feature positions detected by the skin color model are subsequently adjusted by using edge information from the detected facial region along with the proportions of the face. The experiments show that the proposed method guarantees fast and exact processing for face and facial feature point generation and is robust to various lighting conditions and input images.

This work as supported by grant No. (R01-2002-000-00010-0) from Basic Research Program of Korea Science and Engineering (KOSEF).