SAR image despeckling with adaptive sparse representation
Zhenchuan Pang, Guanghui Zhao, Guangming Shi, Fangfang Shen · 2015
SR-based denoising methods have shown promising performance in image denoising. However, Because of the degradation of the noisy image, conventional SR based denoising models may not be accurate enough for the reconstruction of a clean image. Therefore, to reduce the noise corruption, a novel adaptive sparse representation based SAR image despeckling algorithm is proposed in this paper, where the noise component is considered as the coefficient residual, which equals to the difference between the actual image coefficient and the estimated coefficient. By imposing the sparsity constraint on this residual, the noise corruption can be somehow reduced. Furthermore, both the autoregressive model and the nonlocal similarity are incorporated to characterize better the image details. The experimental results demonstrate that the proposed algorithm outperforms other algorithms both subjectively and objectively.