eduzhai > Applied Sciences > Engineering >

Plug-and-Play Image Restoration with Deep Denoiser Prior

  • king
  • (0) Download
  • 20210507
  • Save

... pages left unread,continue reading

Document pages: 14 pages

Abstract: Recent works on plug-and-play image restoration have shown that a denoisercan implicitly serve as the image prior for model-based methods to solve manyinverse problems. Such a property induces considerable advantages forplug-and-play image restoration (e.g., integrating the flexibility ofmodel-based method and effectiveness of learning-based methods) when thedenoiser is discriminatively learned via deep convolutional neural network(CNN) with large modeling capacity. However, while deeper and larger CNN modelsare rapidly gaining popularity, existing plug-and-play image restorationhinders its performance due to the lack of suitable denoiser prior. In order topush the limits of plug-and-play image restoration, we set up a benchmark deepdenoiser prior by training a highly flexible and effective CNN denoiser. Wethen plug the deep denoiser prior as a modular part into a half quadraticsplitting based iterative algorithm to solve various image restorationproblems. We, meanwhile, provide a thorough analysis of parameter setting,intermediate results and empirical convergence to better understand the workingmechanism. Experimental results on three representative image restorationtasks, including deblurring, super-resolution and demosaicing, demonstrate thatthe proposed plug-and-play image restoration with deep denoiser prior not onlysignificantly outperforms other state-of-the-art model-based methods but alsoachieves competitive or even superior performance against state-of-the-artlearning-based methods. The source code is available atthis https URL.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×