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

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.