A sparse representation based super-resolution image reconstruction scheme utilizing dual dictionaries

Super-resolution (SR) image reconstruction is a technique to generate a high resolution (HR) image from several low resolution (LR) images of the same scene, which can improve the visual effect of images or serve as a pre-processing technique. Among various SR image reconstruction schemes, sparse representation based SR image reconstruction schemes have become the current research focus because of their excellent reconstruction quality. In this paper, an SR image reconstruction algorithm based on sparse representation with a redundant dictionary is proposed. In our algorithm, the redundant dictionary and the coding dictionary are trained jointly, and the representation coefficients can be calculated by simply multiplying the input signal by the coding dictionary, which can reduce the computational complexity greatly. The proposed algorithm makes full use of some constraint terms such as the consistence of sparse representation coefficients between the HR image and corresponding LR images, a sparsity prior, autoregressive models and nonlocal means regularization to build up the cost function for SR image reconstruction, and then solves this function based on the iterative shrinkage method to obtain the target HR image. Experimental results demonstrate that the proposed method can achieve great improvement in terms of visual effect, PSNR and SSIM.