denoising empty maps

[tjlane]

When a user denoises a map that is truly empty, it seems to produce huge blobs in the resulting +/-3-sigma Fo-Fo map.

Can we perform simulations that let us understand this behavior, and perhaps identify it? Warning the user would be fantastic.

@alisiafadini

[alisiafadini]

I wonder if we could analyze the map before and after:

Brainstorm
subtract the denoised map from the noisy input. Residuals should resemble random noise. Structured patterns or clumps could indicate artifacts.

I would propose testing skimage.restoration.denoise_tv_chambolle's behavior on 2D images for:

• an empty image
• a slightly noisy image with no other signal

I suspect it's the second case that will be problematic.

[tjlane]

It turns out this is quite problematic, with many users denoising maps with either no or very little signal and then getting stuck trying to interpret the (unhelpful) results. Ideas to catch "low signal" situations:

1. Test for isomorphism using CC vs resolution
2. Check for significant diffmap skew in (non-denoised?) result
3. Compute CC between non-denoised and denoised map, perhaps just in an ROI discovered by the denoising procedure