Segmentation Of SAR Image Based On GMLR And SVMMAR Model

SAR image segmentation is a key technique of automatic target recognition(ATR) and information processing. The existence of speckle noise in SAR imagery make segmentation more difficult than that of optics image. We present several new segmentation methods based on generalized multiresolution likelihood ratio(GMLR) and spatially variant mixture multiscale autoregressive(SVMMAR) model.Firstly, a new expression of multiscale likelihood ratio is obtained, as a result of generation and perfection for the previous algorithm; Secondly, a new multiresolution likelihood ratio test function, generalized multiresolution likelihood ratio, is defined, which cumulates the difference of background and targets at different multiresolution of synthetic aperture radar (SAR) imagery. So the multiresolution likelihood ratio increases the distinction of background and targets. Then we classify each individual pixel based on a test window; Thirdly, an efficient spatially variant mixture multiscale autoregressive(SVMMAR) model is presented. The model is not only capable of describing spatially variant characteristics but also exploits multiscale autoregressive statistical properties of SAR imagery. The performance of the model have no use for denoising preprocessing, while precise segmented results can be obtained, and a kind of method selecting component number quickly at coaser scale is proposed, which can reduce computation amount greatly; Finally, in order to get unsupervised segmentation, a mixture multiscale autoregressive(MMAR) model and an efficient spatially variant mixture multiscale autoregressive (SVMMAR) model are proposed and applied to estimate the parameters of the generalized multiresolution likelihood ratio.The methods proposed avoid some drawbacks that existed in some popular segmentation techniques. Experimental results demonstrate that our methods perform fairly well.