eduzhai > Applied Sciences > Engineering >

Image Super-Resolution with Cross-Scale Non-Local Attention and Exhaustive Self-Exemplars Mining

  • Save

... pages left unread,continue reading

Document pages: 10 pages

Abstract: Deep convolution-based single image super-resolution (SISR) networks embracethe benefits of learning from large-scale external image resources for localrecovery, yet most existing works have ignored the long-range feature-wisesimilarities in natural images. Some recent works have successfully leveragedthis intrinsic feature correlation by exploring non-local attention modules.However, none of the current deep models have studied another inherent propertyof images: cross-scale feature correlation. In this paper, we propose the firstCross-Scale Non-Local (CS-NL) attention module with integration into arecurrent neural network. By combining the new CS-NL prior with local andin-scale non-local priors in a powerful recurrent fusion cell, we can find morecross-scale feature correlations within a single low-resolution (LR) image. Theperformance of SISR is significantly improved by exhaustively integrating allpossible priors. Extensive experiments demonstrate the effectiveness of theproposed CS-NL module by setting new state-of-the-arts on multiple SISRbenchmarks.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×