Energy based Cluster Update (ECU) algorithm proposed by Ralf has its drawback that it is inefficient in sampling and slow in convergence. A new algorithm is proposed to improve the ECU algorithm. It preprocesses the image to merge single pixels to atomic region, thus processes the image on atomic region instead of pixels with the algorithm. It improves the sampling method with Metropolis sampler, which speeds up the convergence of the model. It successfully uses the algorithm on both the static images and video sequences as an application. Experimental results show that the algorithm is more efficient and has higher speed.