eduzhai > Applied Sciences > Engineering >

Dual Adversarial Network Toward Real-world Noise Removal and Noise Generation

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 19 pages

Abstract: Real-world image noise removal is a long-standing yet very challenging taskin computer vision. The success of deep neural network in denoising stimulatesthe research of noise generation, aiming at synthesizing more clean-noisy imagepairs to facilitate the training of deep denoisers. In this work, we propose anovel unified framework to simultaneously deal with the noise removal and noisegeneration tasks. Instead of only inferring the posteriori distribution of thelatent clean image conditioned on the observed noisy image in traditional MAPframework, our proposed method learns the joint distribution of the clean-noisyimage pairs. Specifically, we approximate the joint distribution with twodifferent factorized forms, which can be formulated as a denoiser mapping thenoisy image to the clean one and a generator mapping the clean image to thenoisy one. The learned joint distribution implicitly contains all theinformation between the noisy and clean images, avoiding the necessity ofmanually designing the image priors and noise assumptions as traditional.Besides, the performance of our denoiser can be further improved by augmentingthe original training dataset with the learned generator. Moreover, we proposetwo metrics to assess the quality of the generated noisy image, for which, tothe best of our knowledge, such metrics are firstly proposed along thisresearch line. Extensive experiments have been conducted to demonstrate thesuperiority of our method over the state-of-the-arts both in the real noiseremoval and generation tasks. The training and testing code is available atthis https URL.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×