Abstract: Noise will be unavoidable in image acquisition practice and denoising is a necessary step to recover the image quality. Synthetic Aperture Radar (SAR) images are inherently exaggerated by speckle noise which occurs due to coherent nature of the scattering phenomena. Denoising SAR images aim at removing speckle while preserving image features such as texture, edges and point targets. The mixture of nonlocal grouping and transformed domain filtering has directed the modern denoising techniques. However, this approach makes a tough assumption that image patch itself provides an excellent approximation on the true parameter which leads to bias problem predominantly under serious speckle noise. Another disadvantage is that the generally used patch pre-selection methods cannot efficiently exclude the outliers and damage the edges. In this study, the SAR image is injected with speckle noise and then edge based marker controlled watershed segmentation is applied to identify the homogeneous regions in SAR image. For each region, the neighborhood pixels are identified by using Intensity Coherence Vector (ICV) and are denoised independently by using a local mean filtering. By separating coherent pixels from incoherent pixels, ICV’s provide finer distinctions among blocks; finally, the blocks are aggregated to form the denoised image. The experimental results show that the proposed method outperforms other methods such as patch-based filtering, non-local means, wavelets and classical speckle filters in terms higher signal-to-noise and edge preservation ratios comparatively.