Abstract: Due to different illumination，skin color and racial factors in practical world，normally it is hard to establish a general skin model to detect all the bare skin.In this paper，an adaptive skin modeling method is studied，i.e.，faces are detected by means of AdaBoost algorithm，and skin modeling is performed by using these detected face regions.In order to achieve good modeling performance，the detected face regions are properly reduced，and a special information fusion technology “AND” operation is used in multiple color spaces.This algorithm does not need to consider illumination，skin color and racial factors，and is an adaptive skin modeling method.Experiment shows that this method can effectively solve most of skin detection problems in color images.

Key words: AdaBoost algorithm, adaptive skin color modeling, multiple color space, skin probability map, information fusion

摘要： 客观世界中存在不同的光照、肤色和人种等因素，通常情况下难以建立一个通用的肤色模型进行各种裸露皮肤的检测。研究了一种自适应肤色建模方法，即利用AdaBoost算法检测人脸，通过这些人脸区域进行肤色建模。为了取得较好的肤色建模效果，适当缩小了由AdaBoost算法检测到的人脸区域;利用多颜色空间信息融合技术，即通过选取多个颜色空间的若干颜色分量，计算待检测图像中这些分量的各自SPM（肤色概率图），经过“与”运算融合获得最终检测的肤色区域。该算法不需要考虑光照、肤色和人种等因素，是一种自适应的建模过程。实验表明，该算法可以有效解决绝大多数情况下的彩色图像肤色检测问题。

关键词: AdaBoost算法, 自适应肤色建模, 多颜色空间, 肤色概率图, 信息融合