Background:
I recently decided to trial PixInsight but soon realised that my ageing Core2Duo laptop (2.2GHz, 4GB ram, 32Bit) was having a mild heart attack while attempting to integrate 18megapixel subs (5184x3456 pixels). Integrating just two subs would result in 'out of memory' errors.
Some information on the PI website mentions playing with the integration Buffer size and Stack size for 32bit Windows users to get rid of the 'out-of-memory' errors.
This did the trick on two subs, so I tried to integrate 10 and the error returned. I toyed around a bit again with the settings and found that I could go no further than integrating 4 of the large 5184x3456 pixel images succesfully.
Sadly this would just not do. I have targets where I've taken over 90 subs.
Thoughts of having to spend ~$1500 on a decent current laptop started to invade my mind.
While further reading the PI page above, I saw the example of the effects of image combination (
see here, scroll down a little).
I'd seen examples like this before but this time I got to thinking.... do I really have to integrate
all subs from an imaging session in one hit?
Is there anything stopping me from integrating all my subs in groups of four?
I couldn't think of any reason why it would not work.....
I had a quick dig around the interweb for the answer but came up empty, so decided time spent searching for the answer would be time better spent testing the idea. I got started.
The Test:
Rather than possibly wasting many hours testing an idea on large images that may not work, I lifted the 'original' test image from PI's Integration page(
image #1), opened it in PS and made 64 copies then slapped a 20% Gaussian noise on each of them(
image #2).
I needed an All-At-Once integration to compare results with, so I opened all 64 of them in PI, used the Integration settings from Harry's tutorials, and let rip.
The next thing to do was test the group of four integration.
I opened images 01 to 04 and integrated, the result is
image #3.
This was then saved as 0104.tiff and the process repeated: images 05 to 08 integrated and saved as 0508, again with 09 to 12, 13 to 16 and so on thru to 64.
I now had 16 subs instead of 64.
The next step was to integrate groups of 4 of the above results, so this was done with image images 0104 + 0508 + 0912 + 1316. This was saved as 0116, the next group 1732 and so on.
The result is
image #4. I now had 4 subs instead of 16.
Last step was to integrate the four results from the last step, which is now essentially all 64 subs. The result is
image #5.
Conclusion:
Well I don't know, but it looks pretty good to me. It is certainly good enough to allow me to go ahead and stack 90+ subs in groups of four
Infact, to my eye, the AAO result(All-At-Once) looks a bit rougher than the GOF(Groups-Of-Four) result, but I don't know what to make of this?
See
image #7, AAO on the left, GOF on the right.
Both results were finally opened in PS for a level stretch and this is where I noticed that the histogram for the AAO image could not be stretched without clipping but the GOF image had some latitude at both ends.(
image #6, AAO top) I'm guessing this means GOF has less dynamic range?
All I could do was adjust the midtone of the AAO image to make it's tone look as close to the original image's as possible, while the GOF image had black point adjust and also midtone.
Image #8 - Original and GOF, levels adjusted @ 200%
