Speckle reduction in SAR images using non-local means filter and variational framework

In this study a novel approach which combines the advantages of Total Variation based Sparsity Driven Despeckling with Quadratic Linear (SDD-QL) regularization term and Nonlocal Means is proposed in order to improve the despeckling quality. Both SDDQL and Nonlocal Means have advantages and disadvantages on speckle reduction. SDD-QL performs really well on homogeneous areas and quite fast but deteriorates the texture areas on the image. Nonlocal means preserves the textures and structures in the image but it is really slow and may cause some artifacts on homogeneous areas with high speckle noise. SDD-QL and Nonlocal Means are combined in a single cost function using a texture map as a weighting variable. Texture map is a binary matrix that is extracted from the image. Each value in the texture map indicates whether a pixel is classified as texture or homogeneous. Nonlocal means is applied on texture areas on the image and SDD-QL is applied on homogeneous areas. With combination of the two methods with using the information from the texture map, a better speckle reduction performance is achieved without deterioration of texture areas. Speckle reduction performance of the proposed method is presented using images with Gaussian and speckle noise with low, moderate and high noise levels and real-world SAR images.