基于改进的2-D Otsu方法和YCgCr空间的肤色分割

为了弥补基于固定阈值的肤色分割方法存在的缺陷,在对多种彩色空间和肤色模型进行分析的基础上,提出采用改进的2-D Otsu方法和YCgCr彩色空间进行肤色分割。首先将光照补偿之后的肤色样本图像从RGB彩色空间转换到YCgCr彩色空间,并利用样本图像上的179221个肤色点建立2维高斯模型;进而将待分割的图像进行光照补偿并转换到YCgCr彩色空间,利用已经建立的高斯模型计算图像的肤色相似度,得到肤色相似度图像;最后,结合像素的空间邻域信息,使用改进的2-D Otsu方法对肤色相似度图像进行2值化处理。对这种方法进行了理论分析和实验验证。结果表明,该肤色分割算法有效地克服了使用固定阈值法进行图像分割时缺乏针对性和抗噪性的缺陷,该算法是可行的。

Skin color segmentation based 0n improved 2-D Otsu and YCgCr color space

In order to deal with the disadvantages of fixed-threshold method in skin color segmentation, a new algorithm based on 2-D Otsu and YCgCr color space was proposed after analysis and comparison of different color spaces and skin color models. Firstly, the skin color sample images compensated with light were transferred from RGB to YCgCr color space and the 2-D Gaussian skin color model was established based on the 179221 skin pixels. Secondly, the image to be segmented was light compensated and transferred from RGB to YCgCr color space. Thirdly, skin color similarity degree was computed based on the 2-D Gaussian model and the skin color similarity image was obtained. Finally, an improved 2-D Otsu method was used into segmentation of the skin color similarity image. Theoretical analysis and experimental results show that the new skin color segmentation method is superior to the traditional methods based on a fixed-threshold in pertinence and anti-noise robustness.