dannie125/MM-BSN

How to get the information of noise spatially correlation?

Opened this issue · 2 comments

hyunW3 commented

Thanks for your code and nice work.
I'm currently studying solving denoising task concerning correlation noise.
image

In Figure 2 of your paper, you inspect the noise correlation information of 0228_N.png from SIDD validation.
I'm just wondering how did you get (or derive) information of spatially correlated noise (a) and sizes of different noise area (b).

Could you give me code or details about it to help me understand?

dannie125 commented

Thank you for your interest in our work.
We calculated the spatial connection noise obtained from the connected domain for every noise map.
We defined the pixels number（eg. >25 in x axis） in the connected domain as the area size of spatially correlated noise, and we calculated the propotion of different areas for one image (eg. 0228_N.png).
I hope my explanation has helped you to better understand the situation.

  • 👍1
hyunW3 commented

Thank you for your quick reply,
I'm sorry that i do not completely understand your answer.

  1. The noise map is 'noisy sRGB image' - 'clean (gt) image' ?
  2. How do you obtain connected domain for noise map?
  3. The pixel number in the connected domain means the number of pixel which is in connected domain?

The sample code would help me to better understand.
Thank you