A Unified Learning Framework for Single Image Super-Resolution

摘要: It has been widely acknowledged that learning- and reconstruction-based super-resolution (SR) methods are effective to generate a high-resolution (HR) image from single low-resolution (LR) input. However, learning-based prone introduce unexpected details into resultant HR images. Although do not obvious artifacts, they tend blur fine end up with unnatural results. In this paper, we propose new SR framework seamlessly integrates for to: 1) avoid artifacts introduced by 2) restore the missing high-frequency smoothed SR. This integrated learns dictionary LR input instead of external images hallucinate details, embeds nonlocal means filter in enhance edges suppress gradually magnifies desired high-quality result. We demonstrate both visually quantitatively proposed produces better results than previous literature.