channel combination and image merge

[rxdeath]

hello pi friends

i'm attempting to work with pixinsight and am having a little bit of difficulty.  I am trying to make a full moon image from perigee last nite.   i took 5 movies @ 12 seconds each for r/g/b, then used my normal (for solar) process with autostakkert to combine them into 5 subs.  because of the large moon and i used a 2x powermate to fill up my chip, i had to do a 'top' and 'bottom' exposure set.

then following some tutorials and what i've learned working on my (pitiful) pleiades image, i made a master using star alignment which worked, then integrated to provide a solid image for each channel.  i did this for both top and bottom with no obvious problems.  then i registered each of the 3 masters, and did a channel combination to make a color image.  i happened to do the 'top' first and gota great color lunar image.  then i did the bottom and it just refuses to 'turn color'.  the image that comes up has no overlap/bleed, but it doesn't do what the 'bottom' did and turn into a nice color photo, it's just a greyscale that doesn't look much different from the source files.  i'm hoping someone can tell me what could cause this because the workflow is exactly the same for top and bottom, but it just won't go.

as a 2 parter since i'm getting some help, is there a good direction to look for-
a lunar processing tutorial  since i know pixinsight is usually used for star fields and such, but i'm sure it'd been done

a mosic processing tutorial since ideally i'll get this sorted out and will have to stitch them together, i'd rather not use a different program at the very end since i've done almost everything in pixinsight after i got it stacked. lightvortex has one but i don't think it will apply well to a lunar image.

thanks for any help, i'm so close i don't want to turn back.  i can post a zip with all the master files if it helps, its 233MB so i'd probably just post it to my webserver for download.  i bet it's something simple because i'm still a beginner, i use nothing but star alignment, subframe selector, image integration, channel combination.  its a pretty big image at 4530x3500 i used an asi1600 for acquisition, each master frame is about 64mb and the successful top color frame is ~180mb.

[rxdeath]

figured i'd be proactive


www.rxdeath.com/moonparts.rar

these are the 3 registered integrated masters for top and bottom, if you do a channelcombo on top, it works as expected, bottom yields a greyscale image as described above

[RickS]

ChannelCombination gives a colour image in both cases but they don't match well.  If you look at the results with Statistics or check the histogram then you'll see that the R, B and G channels are different in both combined images.  I don't have time to look any further right now, but will play with the data a bit more later today if I get a chance...

Cheers,
Rick.

[rxdeath]

oh ok, thank you for your time, i really appreciate it.  the top image is consistent with colorations i get using other programs.  the bottom image doesn't match what i normally get, of course the moon is ok in greyscale, but that nice color makes it pop i think.  the exposures are all similar, i did a sequence of 5 for the 3 tops, then a sequence of 5 for the 3 bottoms.  the around about 3 minutes apart time wise, and the conditions in the sky were pretty consistent.  the histograms saturation is pretty close in the video data, so i'm not sure why such a drastic difference.  i guess the saying isn't close counts in horseshoes, hand grenades, and astrophotography.

PI is wonderful the more i use it the more i realize its by far the most powerful processor for ap images in existence, however the learning curve is pretty steep.  most users seem to be experts at both computer science and imaging so it's a little hard to follow for new people.

thanks for a helping a newb crutch through his project with any time you can spare, i took these right around the minute of 100% moon ~18:24 local time so its a true 'full moon' or at least as close as i've ever gotten, that coupled with perigee makes it a neat picture for me.

ps- i noticed that the blue line from bottom was pretty close to the red on top, so i tried switching the channels to see if it would make a difference or get it closer, but still no love

[RickS]

I placed a small preview in the same position on all the images and looked at the stats (median value).  The match between top and bottom in red and green is fairly close (about 1% different in red, 3% in green).  For the blue, the median in the top is roughly 10% greater than the same area on the bottom.  So, looks like the problem is earlier in the process.

You can tweak the data a little and get a closer match, e.g. apply the PixelMath expression "$T*1.1" to the bottom blue.  It's not perfect but it's an improvement.

Cheers,
Rick.

[rxdeath]

thanks i  think i understand why its inconsistent and i will try to fix it as you said.

if you have a moment, i'd really appreciate a pm or something of what you did to diagnose the problem in a click by click run down, there are so many options in PI, it takes me forever to figure things out, and if i know how to do this, i can add it to my toolbelt.  it sounds like a basic task that will be needed frequently in my future endeavors.  i did use the histogram and noticed the blue was pretty high compared between top and bottom. but your analysis and 'preview' far surpassed mine.

also i've used a little pixel math to do after-the-fact binning, but not for anything like this, do you have a good reference i can scan to learn more about it?

sorry to ask for the handholding, if there is a good 'instruction book' i have yet to find it, and i'm happy to do the work myself, its just overwhelming to find the right reference

as they say, give a man a fire and he's warm for a day; light him on fire, he's warm for the rest of his life

thanks again for your time

[pfile]

Inside Pixinsight by Warren A. Keller

rob

[RickS]

I second Rob's recommendation of the "Inside PixInsight" book.

if you have a moment, i'd really appreciate a pm or something of what you did to diagnose the problem in a click by click run down, there are so many options in PI, it takes me forever to figure things out, and if i know how to do this, i can add it to my toolbelt.  it sounds like a basic task that will be needed frequently in my future endeavors.  i did use the histogram and noticed the blue was pretty high compared between top and bottom. but your analysis and 'preview' far surpassed mine.

I created a new preview on one of the "top" images (Alt-N and click/drag to define preview.)  I then displayed the preview (click on its name on left hand bar) and then dragged it to the left hand bar of the two other "top" images.  This creates identical previews on all three "top" images.

Then I copied one of the previews to a "bottom" image, went into edit preview mode (Alt-E) and dragged the preview to the same point on the moon matching the "top" images.  I did this by eye but zoomed the images up to 8:1 to try and do it fairly accurately.

Once I had matching previews on all 6 images I just looked at the stats for each using the Statistics process (selected the previews one at a time in the view selector at the top of the Statistics dialog.)  I noted and compared the median value for each preview.

Cheers,
Rick.

[rxdeath]

you guys are great, thanks so much.  i'm going to try this tomorrow, and see what happens, then work on the proposed pixel math to fix it up.

also, i bought the book you recommended, so i will put my nose to the bookstone

[RickS]

Good luck and let us know if we can help...

[rxdeath]

ok, i've been through your process and see what you did to evaluate the image, then replicated it on my end and see what you did.  of course i wouldn't know to do that, but i see what happened, and i also see the variance between the blue images.   i of course know median but i'm not sure why you chose that over values to make your comparison, and why that is important. i would guess it shows that the one image is going to be different because the midpoint is higher so it's going to be a decent amount brighter, and have more 'weight' going into the color image which is why i have a blue tint to the image.

so any insight into the mental processes that went into it is appreciated, but really i have this question.  did you match bottom to top because i asked for that or because it's what is 'right'?  how do i know which is closer to true values, i used the histogram to compare the 2 color images and they are obviously different, but i'm not sure what of that has to do with the exposures i got that night (the histograms vary between 77 and 100 on most of them so i had atmosphere/weather issues) and what part has to do with the true color balance of the moon.

i'm excited to get through this book and understand more about pixinsight, but my general photo knowledge also has a long way to go so hopefully i can figure out what i want to do to my pictures, and then use pixinsight to do it.   maybe i need to buy astrophotography for dummies

your pixel math expression nailed it, the panorama it made was very nice and matches well enough no one i showed it to had anything to say but wow.  that's the part i'm most excited to learn about, the pixelmath of PI seems to be incredibly powerful and can make just about anything happen if you know the right expressions.

thanks for teaching me how to evaluate the balance of images in my RGB mosaics!

[RickS]

I used the median as it's a robust way to measure a "typical" value.  I could have used the mean but it is influence more by very big or very small values.

Matching bottom and top was what you asked for but also was the obvious thing to do.  It's a common problem for mosaic panels not to match and I have dealt with it before with images of deep sky objects.

I'm used to doing this sort of stuff but it's certainly possible to learn even if you don't have a strong background in maths and computing.  Just read Warren's book, dig through PI forum posts, etc.  You'll soon pick it up.  It helps to have some natural talent for troubleshooting.  That's how I see image processing.  It's repeatedly finding and fixing problems in the image until it looks great 

Cheers,
Rick.

[rxdeath]

ah what i meant is why didn't you propose doing the pixel math to the 'top' image with a .9 instead?   is my blueish tint 'correct' or how do i know i'm as close to real as possible?  i see a multitude of different color tinges to the moon and the ones with the brown in them seem to me the most desirable, but i don't see that blue tinge so i'm wondering if my bottoms were closer to correct then the top and i just had a little bit more in the histograms for the blue on those due to clearer viewing or other issues.

[RickS]

ah what i meant is why didn't you propose doing the pixel math to the 'top' image with a .9 instead?   is my blueish tint 'correct' or how do i know i'm as close to real as possible?  i see a multitude of different color tinges to the moon and the ones with the brown in them seem to me the most desirable, but i don't see that blue tinge so i'm wondering if my bottoms were closer to correct then the top and i just had a little bit more in the histograms for the blue on those due to clearer viewing or other issues.

Sorry, I misunderstood - I tried to match the top because that was the one you thought was good.  I have never done any lunar work and don't have an informed view as to what colour it should be.  When imaging DSOs I normally use the stars for colour calibration but that's not an option here.  I reckon it would be interesting to try the ColorCalibration process on your data with the whole moon as a white reference but for advice on truly accurate lunar colour you'd need to talk to someone else

Cheers,
Rick.