高斯混合函数区域匹配引导的Level Set纹理图像分割

Texture Image Segmentation Using Level Set Function Evolved by Gaussian Mixture Model

基于高斯混合模型颜色匹配及多尺度图像增强,文中提出了有效的边缘停止函数用于引导levelset函数演化,有效地解决了纹理图像的分割问题.文中首先提出基于高斯混合模型颜色分布的边缘停止函数,通过计算levelset演化窄带区域与用户给定交互区域的相似性,根据其相似性来引导levelset快速演化;然后,提出一个定义在多尺度图像梯度上的边缘停止函数,使得levelset能精确地分割出图像的边缘;最后,结合以上两种边缘停止函数的优点,提出一个边缘停止函数的混合模型,根据图像颜色、边缘特征自适应地引导levelset函数演化.实验结果表明,文中提出的算法不仅能有效地检测出纹理目标区域,同时需要计算出纹理区域精确、光滑的边界.

This paper presents an effective level set texture image segmentation approach that incorporates GMM （Gaussian Mixture Model） and multi-scale image enhancement techniques. First, the authors construct a new edge stop function based on GMM to guide the evolution of the level set function over the regions with similar texture, and the edge stop function is computed according to the similarity between the narrow brand pixels near the zero level set and the specified regions selected by the user. Then, to accurately detect the boundary of the texture image, a multi-scale edge stop function is defined on the gradient domain of the image, thus an accurate boundary result can be achieved. Finally, these two methods are combined to develop a mixing edge stop function, which forces the level set to evolve adaptively based on the texture and gradient. As the results show, the new approach is effective for texture image segmentation and works well to detect the accurate and smooth boundaries of the object.