Abstract: 2-D entropic thresholding method is a common image segmentation technology，it is suitable for the image segmentation of the object and background occupying the vast majority region of 2-D histogram，but this approach neglects the information of edge regions to the results of the segmentation.To preserve the image of the edge regions，this paper improves the select method of the mean of the histogram，uses different shape templates，achieves mean and variance of each pixel point，selects the mean of smallest variance template to struct two-dimensional histogram.Experimental results show that this method can obtain better segmentation results and denoising effect.