Spatially Varying Regularization of Image Sequences Super-Resolution

This paper presents a spatially varying super-resolution approach that estimates a high-resolution image from the low-resolution image sequences and better removes Gaussian additive noise with high variance. Firstly, a spatially varying functional in terms of local mean residual is used to weight each low-resolution channel. Secondly, a newly adaptive regularization functional based on the spatially varying residual is determined within each low-resolution channel instead of the overall regularization parameter, which balances the prior term and fidelity residual term at each iteration. Experimental results indicate the obvious performance improvement in both PSNR and visual effect compared to non-channel-weighted method and overall-channel-weighted method.