eduzhai > Applied Sciences > Engineering >

Improving Blind Spot Denoising for Microscopy

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 15 pages

Abstract: Many microscopy applications are limited by the total amount of usable lightand are consequently challenged by the resulting levels of noise in theacquired images. This problem is often addressed via (supervised) deep learningbased denoising. Recently, by making assumptions about the noise statistics,self-supervised methods have emerged. Such methods are trained directly on theimages that are to be denoised and do not require additional paired trainingdata. While achieving remarkable results, self-supervised methods can producehigh-frequency artifacts and achieve inferior results compared to supervisedapproaches. Here we present a novel way to improve the quality ofself-supervised denoising. Considering that light microscopy images are usuallydiffraction-limited, we propose to include this knowledge in the denoisingprocess. We assume the clean image to be the result of a convolution with apoint spread function (PSF) and explicitly include this operation at the end ofour neural network. As a consequence, we are able to eliminate high-frequencyartifacts and achieve self-supervised results that are very close to the onesachieved with traditional supervised methods.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×