Here's what I'm trying:
1. Just do normal image integration of all 357 frames using GESD test - doesn't work great but faint borders can be seen.
2. Do drizzle integration of all 357 frames using GESD test - seems to work better but can still see borders. It's not bad at all actually and may work and is so far the best I've gotten with the best SNR.
3. Try to do AN on 1-2, which doesn't work due to the dark corners. I could try LocalNormalization. Should I?
4. Do 1-2 separately on the set of 153 and 204 frames and then combine using HDRComposition, GradientHDRComposition, and GMM. Previously when I experimented with 1/4 of the data, I found GHDRC worked like a charm and produced the best SNR after combination but now it doesn't work at all and GMM does work with a marginal SNR improvement.
Thoughts? I really would've thought AN would've worked but perhaps I should try LN.
You may ask why not just do 4 which I've tested with a smaller number of frames and I think GHDRC really worked well. But I'm not sure it's the optimal path to take. 90% of the pixels are common in all 357 frames so why not take advantage of this data? The statistics would be more robust with 357 frames instead of half the frames, no?
Thanks!
--Ram