Despeckling SAR Images with an Adaptive Bilateral Filter

In this paper, a bilateral filter is proposed for despeckling Synthetic Aperture Radar (SAR) images with an adaptive estimation of the domain and range parameter. The method applies a block-based approach to perform Laplacian Data masking and gradient operation for each pixel in the image utilizing its local neighbors. The range parameter is then adapted to the noise level estimated from the corresponding edge removed image. On the other hand, the domain parameter is adapted according to coefficient of variation of local neighboring pixels of the block. Experiments performed on synthetically-speckled images show that the proposed method outperforms several existing methods for despeckling in terms of the peak signal-to-noise ratio, structural similarity and edge preservation indices and visual quality. For real SAR images, it preserves the important details of the image while suppresses the speckle in homogeneous regions considerably.