A non-local means based vectorial total variational model for multichannel SAR image denoising

A non-local vectorial total variational model is proposed for multichannel Synthetic Aperture Radar (SAR) image denoising. By introducing non-local means to the vectorial total variational model, it is found that the new non-local vectorial total variational model is well performed on denoising while preserving the fine structures of the multichannel SAR images. The energy functional of non-local gradient formed vectorial total variation is well constructed. Followed, the discrete version of this model is designed for constructing the fixed point iteration to solve the proposed model. The new algorithm is implemented on multipolarization RADARSAT-2 images. Result shows that the non-local vectorial total variational model fits well. The convergence is proved as well.