C21. Super Resolution Using Dictionary Technique

This paper addresses the problem of generating a super resolution (SR) image from a degraded input image, using a hybrid algorithm called Kernel-Local Learned Dictionary (K-LLD) to form dictionary from low resolution (LR) image to produce high resolution (HR) one, a patch-based, locally adaptive denoising method based on clustering the given degraded image into regions of a like geometric structure. It will show the effectiveness of sparsity as a prior for regularizing, where we utilize as features the local weight functions derived from steering kernel regression, to effectively perform such clustering. Next, we model each region (or cluster) which may not be spatially contiguous by "learning" a best basis describing the patches within that cluster. This learned basis (or "dictionary") is then employed to optimally estimate the underlying pixel values using a kernel regression framework. In addition, this paper illustrates the overall algorithm capabilities with several examples. Sparse K-SVD algorithm is applied for optimization to speed up sparse coding. Comparison with sparse coding method shows that sparse dictionary is more compact and effective.