New Approach for SAR Imagery Segmentation Based on Multiscale Likelihood Ratio Test
一种新的基于多尺度似然比检验的SAR图像分割方法

This algorithm exploits characteristic variations in speckle pattern as image resolution is varied from coarse to fine, and so does Markov property among scales. A new expression of multiscale likelihood ratio is obtained. Algorithm fuses residuals produced by MAR model and gray value information of coarse scale image. It increases the distinction of different targets, in the meanwhile, segmentation is precise. The use of window in the pixel-by-pixel classification makes the method insensitive to noise and robust. Experimental results demonstrate that the method performs fairly well.