Image segmentation using histogram fitting and spatial information

In this paper, we introduce a novel unsupervised segmentation method using a histogram fitting method to find out the optimal histogram clustering based on multi Gaussian models. The fitting problem is performed via the trust region reflective Newton method to minimize a predefined cost function. The histogram clustering is the global information describing the probability of a given gray value belonging to a category. Together with the consideration of the spatial information, the image segmentation is performed. We demonstrate some applications on medical images such as brain CT and MRI.