鲁棒的肤色检测方法

Robust method for skin color detection

肤色检测常用的方法有颜色空间阈值法,肤色统计模型和基于分割的方法。然而,上述方法对于光照不好的图片往往得不到很好的效果,而且在背景颜色与肤色相近时,往往误判背景为肤色。为此,提出了一个综合的方法,它对图片光照和背景颜色具有鲁棒性:先作图像预处理,通过阈值法选出肤色像素,再对肤色密度分布进行估计,最后使用改进的分水岭算法进行肤色区域生长。实验结果表明,该方法可以有效降低背景颜色的干扰,提高了肤色轮廓的精确度。

Many commonly used methods of skin detection are proposed including threshold in color space, skin color statistical modeling and segmentation-based method. However, those methods’performances usually decrease in poor illumination conditions and they often misjudge some parts of the background as skin region when their colors are similar to skin colors. Therefore, a comprehensive method is proposed that is robust to the illumination conditions and background colors. It has mainly two steps. At the first step the image is pretreated, and then selects some skin seeds by threshold to estimate the skin colors density distribution. Finally an improved watershed algorithm is used to expand the skin color regions based on the above density distribution. Experimental results show that the method effectively reduce the interference of the background color to enhance the color contour accuracy.