A Novel SAR Images Restoration Using Multiscale SVR

Synthetic aperture radar (SAR) has become one of the most powerful observation tools in the studies of natural environments and Earth resources. However the granular appearance of speckle noise in synthetic aperture radar imagery makes it very difficult to visually and automatically interpret information of SAR data. In this paper, according to the inherent speckle property of SAR image, we proposed a multiscale restoration algorithm by fusing the wavelet coefficients manipulation technique with support vector regression. The kernel parameter was used respectively in the different scale. For preserving sharp edges information, in our algorithm the shrinkage strategy is to compare the estimated value with the original coefficient value of that pixel using the absolute deviation of them. We define a rule for modifying wavelet coefficients based on support vector regression (SVR). Real SAR images are used to evaluate the restoration performance of our proposed algorithm along with another wavelet-based restoration algorithm, as well as the Lee speckle filter. Experimental results show that the proposed method outperforms standard wavelet restoration techniques.