Comparison of Transform Domain based SAR Despeckling Techniques

Objectives: Synthetic Aperture RADAR (SAR) is a satellite imaging technology which is affected by speckle noise having granular pattern. Speckle is multiplicative noise and occurs due to the interference of the signal with the backscattered echoes. Methods/Analysis: Speckle degrades the image quality and makes further segmentation and classification of images difficult. Despeckling can be done in spatial and transform domain. In this paper the various transform domain despeckling techniques like wavelet, shearlet, contourlet and curvelet are compared. The results are analyzed using performance parameter like ECF, SSIM and ENL. Findings: Comparison of the various methods is done by using synthetic images and real images. ECF and SSIM are used to evaluate synthetic images and ENL is used for real images. In the case of real images the ENL value is highest for curvelet compared to wavelet, shearlet and contourlet. In the case of synthetic images SSIM and ECF value is high for curvelet compared to other methods. ENL value is lowest for shearlet transform whereas SSIM and ECF values are lowest for contourlet transform. From the results, it can be concluded that curvelet outperforms all the other methods. Novelty/ Improvements: Transform Domain techniques has got wide spread applications in the field of denoising, segmentation and classification. Using curvelet for despeckling can improve the further segmentation and classification process.