eduzhai > Applied Sciences > Engineering >

Efficient Blind-Spot Neural Network Architecture for Image Denoising

  • king
  • (0) Download
  • 20210507
  • Save

... pages left unread,continue reading

Document pages: 2 pages

Abstract: Image denoising is an essential tool in computational photography. Standarddenoising techniques, which use deep neural networks at their core, requirepairs of clean and noisy images for its training. If we do not possess theclean samples, we can use blind-spot neural network architectures, whichestimate the pixel value based on the neighbouring pixels only. These networksthus allow training on noisy images directly, as they by-design avoid trivialsolutions. Nowadays, the blind-spot is mostly achieved using shiftedconvolutions or serialization. We propose a novel fully convolutional networkarchitecture that uses dilations to achieve the blind-spot property. Ournetwork improves the performance over the prior work and achievesstate-of-the-art results on established datasets.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×