Strictly speaking this is not about free software per se, but image processing in general and there are a lot of smart people here so I thought I would ask here anyway. If anywhere during reading this post or watching the linked video you experience feelings of mental or physical discomfort, feel free to mentally substitute “Microsoft Excel” with “LibreOffice Calc”.
This post was prompted by this video: https://www.youtube.com/watch?v=bkQJdaGGVM8
TL;DW: An attempt to performe HDR image merging in a spreadsheet application, using convoluted matrix pseudo-inversion.
After watching it, I was left with a couple of questions that I hope you can help me with.
- If you shoot raw, this whole ‘response curve estimation’ shouldn’t be necessary, right? Your sensor data should scale more or less linearly with the amount of light on your sensor and you can ‘just’ merge the data without problems
- If I understand correctly. his response curve is a combination of the jpeg gamma curve, and the tone/contrast curve used by the camera’s jpeg engine. The first thing you do before messing with your data would be compensating for the gamma curve to get the ‘real’ image data.
- Around the 4:00 mark he shows a diagram of what the curve would look like. We see that for twice the captured light, the number is more than twice as high. Wouldn’t this work the other way round for gamma encoded files? I believe that should compress highlights and raise shadows, a bit like a log colour profile for video.
In any case, his methods seem unnecessarily convoluted (I am aware he is basing it on a paper from '97). What is your take on this / am I understanding this correctly?
edit: this is the paper, for those interested: https://people.eecs.berkeley.edu/~malik/papers/debevec-malik97.pdf