Image post-processing and alignment quality degradation

Development Asset Creation
,
1

Hello

I am currently working on a pretty large data set where a lot of images are taken with a strong backlighting and are very underexposed.

As a lot of images are very dark (several hundreds), I am trying to batch process them in photoshop to see the impact on the mesh quality and details

With the images processed, I am able to recover more details in the mesh, also more triangles. But i noticed that the more the images are enhanced, the more the alignment statistics deteriorates (point count and total projections decrease, mean and median error increased). The pixel error on control points also seems to be bigger.
It seems a bit counterintuitive to me.

note that I am only woking with the enhanced images here.

Any technical explaination on why this is happening ?

Thanks !

2

Hello,

By changing the lighting of your images you are revealing more information, but you are also creating different lighting in your images. So even though you are increasing the amount of detail that can be seen, you are also increasing the space for errors to show up. Especially if by modifying images some colors are being compressed.