CS-544 Optimization in Computer Vision

MP 2

 

 

Compare various total variation denoising (TVD) strategies on 64 x 64 intensity images.

  1. Implement TVD for 64 x 64 intensity images using the alternating direction method of multipliers, as sketched in class. Evaluate the denoising effectiveness of your method by (a) adding IID normal noise to an image (b) using your method to remove this noise (c) then computing the root mean square error (RMSE) between the original image and the denoised version.
  2. Apply your method to color images by denoising R, G and B independently. How well does this work? why?
  3. Compare your TVD implementation to an open source implementation. Which is faster? which is better?

This is a broad MP intended to be educational. I will grade on quality of experimental concept and of argument from data. You may find the account at this link helpful; it is by Nikola Janjušević. Submit PDF's on canvas 28 Feb 2024. Submission will be by Canvas - details shortly.

TVD has now been incorporated into network layers by Raymond Yeh and others; look here for details, also very nice discussion of recent stuff for fast algorithms.