A super-resolution method based on local sparse and global gradient

Super-resolution methods based on sparse easily lead to over-smoothing at the edges of reconstructed image. A novel super-resolution method based on local sparse and global gradient is proposed to solve the problem. First, it represents the input low-resolution (LR) image patches with sparse coefficients and LR over-complete dictionary. Then it maps the coefficients to high resolution (HR) over-complete dictionary and reconstructs the HR texture. At last, it enhances the main edge using global natural image statistics´ prior information and merges it together with the texture. By using the local sparse representation and global gradient transformation, it can obtain the result image with clean texture and clear edge. Experimental results validate the proposed method, both in subjective and objective quality.