Image Super-Resolution using a Hybrid Scheme with DCT Interpolation and Sparse Representation Method

Learning-based Super-Resolution methods have attracted much interest in recent years in many signal and image processing tasks. In this paper, we present an algorithm for single image super-resolution that use discrete cosine transform (DCT) interpolation and sparse learning-based super-resolution method. The input LR image is interpolated using both DCT interpolation and bicubic interpolation methods. The patches of bicubic interpolated image, undergoes a process sparse coding using OMP algorithm and training using k-SVD algorithm. The obtained sparse coefficients are multiplied with high-resolution dictionary generated in the training phase, resulting in the intermediate HR image. The final HR image is obtained by adding the DCT interpolated image and intermediate HR image. The experimental results demonstrate the effectiveness of the method proposed in terms of PSNR, SSIM and visual quality