A multiscale kernel approach to speckle suppression for SAR images

Coherent speckle noise of SAR images decreases the image quality seriously, and makes the image difficult to be understood. The recently developed multiscale kernel (MSK) is a truly mesh-free method and can handle unstructured data with noise in any dimension. In this paper, a new speckle reduction method for SAR images based on MSK is presented. Firstly, the feasibility and rationality of filtering method based on MSK is proved by the analysis of decomposition and experiment for one-dimension signal denoising. Then the SAR image is regarded as two-dimension continuous signal and is multiscale decomposed based on MSK. Finally, the decomposed image is filtered by a threshold strategy and a inversed operator is applied to recover the image. Experimental results show that the presented method can both suppress the speckle noise and keep the edges in SAR images, the performance is better than that of the traditional filter methods.