Abstract: A new despeckling scheme for synthetic aperture radar (SAR) images is proposed based on the adaptive shrinkage principle in the nonsubsampled contourlet transform (NSCT) domain. First, the statistical distributions of signal and speckle noise coefficient magnitudes in the NSCT-domain image subbands are effectively approximated by Gamma distributions and exponential distributions, respectively, which can make the shrinkage factor well adapt to the high redundancy of NSCT image subbands. A new set of directional neighborhood models is then proposed to calculate the prior ratio, making the shrinkage factor well adapt to the flexible directionality of NSCT-domain image subbands, thus enhancing the coefficient shrinkage performance. The experimental results on a real SAR image demonstrate that the proposed despeckling scheme can preserve the details while speckle noise is reduced. Compared with several classical despeckling methods, the new scheme has better edge preservation performance and backscattering coefficient preservation performance.

Key words: SAR image, despeckling, nonsubsampled contourlet transform, adaptive shrinkage, directional neighborhood system