肤色检测技术综述

肤色检测在人脸和手势识别与跟踪、Web图像内容过滤、数据库或因特网中的人物检索和医疗诊断等方面有广泛应用，文中通过分别介绍基于统计和基于物理的两类肤色检测技术，较全面地综述了肤色检测技术，其中对颜色空间选择、静、动态肤色建模方法、肤色反射模型和肤色波谱特性等肤色检测重要环节做了分析，明确了选择颜色空间与特征提取和分类方法的联系，强调了研究肤色波谱特征对基于物理的肤色检测技术的重要性，最后探讨了肤色检测的技术难题和发展趋势。

A Survey of Skin Color Detection

Skin color detection has many applications in tasks like detecting and tracking human faces and gestures, filtering web image contents, retrieving people in databases and Internet, and diagnosing diseases. This paper comprehensively surveys the skin color detection techniques by dividing them into statistics-based and physics based approaches. Important aspects of skin color detection are discussed, including color space selection, static and dynamic skin color modeling, skin reflection models, and visual and infrared skin spectrum. Based on the discussion, the color space selection is related to feature exaction and classification methods. The selection of color space is actually the selection of a feature base for the classification;and features are not independent from the classifier, so how well they cooperate can affect the overall performance of the classification system. Therefore, it is of little significance to discuss the optimal color space for skin color detection without considering the methods of skin color modeling and the principle and implementation of the classifier. A successful dynamic skin color model should be able to work well under varying illumination, which is also the precondition for a skin color detection system to work properly outdoors. Physics-based skin color detection should study skin reflection model or skin spectrum, the combination of reflected and emitted data will essentially improve the performance of skin color detection system. The major challenge of skin color detection techniques is how to deal properly with different illumination conditions and complex background. The further research directions of the skin color detection techniques include intercrossing related disciplines, fusing related algorithms and combining related features. It is a trend to integrate physics-based techniques in statistics-based skin detection techniques in high end applications.