基于高斯肤色模型和图像分割的人脸检测

提出了一种基于HSV、YCgCr颜色空间和图像区域分割的人脸检测方法.首先,将规范化的rgb肤色转化到HSV、YCgCr颜色空间,对向量（H,Cg,Cr）T进行统计,得到肤色的三维高斯概率模型.然后,根据该模型计算待检图像的肤色相似概率,采用最大类间方差的方法计算二值化阈值,根据阈值将相似度图像二值化.为了抵消人脸区域亮度变化和阴影的影响,以二值图像为模板,在原待检图像的饱和图中根据3像素×3像素区域的平均R、G值进行图像区域分割.在分割得到的区域中,根据人脸特征检测出图像中的人脸.实验表明,提出的方法能在复杂背景的情况下检测出人脸图像.

Face Detection Based on Skin Color Gaussian Model and Image Segmentation

Based on HSV and YCgCr color space as well as image segmentation,an approach for face detection was put forward.First,the normalized rgb color was converted into HSV and YCgCr color space.Meanwhile the 3D Gaussian model was obtained by statistics of the vector（H,Cg,Cr）T.According to this model,the image similarity of skin color that is ready for being inspected could be calculated,and binary threshold could be attained through Otsu method.As a result,the threshold could binary the similarity image.In the application of binary image,the saturate image was segmented on the basis of the R,G component obtained by the 3 pixel×3 pixel region so as to offset the effect of brightness variation and shadow.In the areas of image segmentation,face was detected from the image by facial features.Experiment shows that the method can detect the image of face accurately in complicated background.