Image Segmentation Using Improved Potts Model
Xiangrong Wang, Jieyu Zhao · 2008
The classical Potts model is a powerful tool for image segmentation but the drawback of the model is its slow convergence. The main reason for this is that there exists a critical slowing down process at phase transitions. To overcome the drawback of the Potts model, an image segmentation method based on the ECU (energy based cluster update) algorithm according to the characteristics of image segmentation is developed. Firstly, with merging single pixels into atomic regions and atomic region instead of pixels, the image is preprocessed and segmented, thus we segment the image using atomic region instead of pixels. Secondly, the Metropolis sampler is adopted to speed up the sampling and the convergence of the model. Finally, the algorithm is successfully applied to segment both the static images and video sequence images. Experimental results show that the proposed method is robust and quite applicable.