基于新颜色空间YCgCr的人脸区域初定位

在RGB颜色空间采用了颜色平衡方法对发生色彩偏移的输入图像进行颜色校正；在新颜色空间YCgCr上建立了亮度和Cg-Cr色度查找表联合的肤色模型，对肤色区域进行检测；引入了有效的预处理技术，进一步去除肤色分割后的二值图像中的部分非人脸区域，减少人脸定位的搜索区域。最后在两个图像测试集上进行了实验比较，实验结果表明，该肤色模型可以有效地从复杂背景的彩色图像中检测出肤色区域，光照条件适应性好，且引入的预处理技术在保证漏检率低的前提下，能够去除大部分非人脸区域。

First Face-region Location Based on a Novel Color Space YCgCr

A color balance method is introduced to correct the image color bias of the input images in the RGB color space. A joint skin-color model of luminance and Cg-Cr chroma lookup table is constructed in the new color space YCgCr to detect skin-color regions in the images. Efficient pre-processing techniques are proposed in order to further remove some non-face regions in the skin-color segmentation results, to reduce the searching regions for face location. Finally, some experiments are done on two testing image sets, experimental results show that this skin-color model is effective to detect skin-color regions from the color images with complex backgrounds, also has more robust illumination adaption, and the proposed pre-processing techniques can remove more non-face regions with the guarantee of lower dismissal rates.