Single Image Super-resolution Based on Gradient Profile Prior and Nonlocal Self-similarity Feature

Image super-resolution has received great attention in recent years. In order to produce a high-quality HR image with minimal artifacts, we propose a new image super-resolution method. We propose a diffusion function to refine the gradient directions along the edges. Based on the neighboring gradient profiles, a GPS optimization function is devised to make the estimated sharpness more accurate. In order to break the limitations of traditional non-local self-similarity method, we propose a new non-fixed search method to search for non-local self-similarity image patches. Besides, gradient profile prior is used for suppressing the ringing artifacts effectively. A new image reconstruction framework is designed by combining gradient profile prior and non-local self-similarity prior. Finally, we propose a high-pass filter function to get the high-frequency components, which then enhance the image quality and edge details by shock filter. The experimental results demonstrate that the new algorithm surpasses the previous state-of-the-art methods, in both visual quality and PSNR /SSIM/IFC performance.