Locally Adaptive De-Speckling of SAR Image using GCV Thresholding in Directionlet Domain

Speckle noise usually occurs in Synthetic Aperture Radar (SAR) images due to coherent radiation. Speckle reduction is a mandatory step prior to the processing of SAR images. Here a novel de-speckling scheme is presented which is in line with the wavelet transform based schemes with several modifications due to the implementation of directionlet transform. As in any transform based de-speckling schemes, the directionlet based de-speckling also involves mainly three steps. First transform the input image using an orthogonal transform, then threshold the transform coefficients using a non-linear algorithm and finally reconstruct the image using the modified coefficients. The effectiveness of a despeckling algorithm basically depends on two factorsone is the efficient representation of the image to be despeckled using a local, directional and multi resolution expansion and second is the efficient computation of an optimal threshold. Here the first requirement is met by using a locally adaptive directionlet transform and the second by optimal threshold computation using Generalized Cross Validation (GCV) technique. Experimental results on simulated and actual SAR images show that minimizing GCV in the directionlet domain results in better de-speckling performance when compared to minimizing it in the wavelet domain. The proposed scheme outperforms many of the traditional de-speckling schemes in terms of numeric and perceptual quality.