Data for Processing

[sreilly]

I've been working on learning PI these past weeks and have watched Harry's videos and read most of the printed tutorials. All are very helpful but ultimately lead to other questions or clarifcations. I've been using my data on M95 which consist of 7 images each of red, green and blue for 15 minutes each and binned 2x2. The luminance data is 29 images each 15 minutes and binned 1x1. The equipment is listed in my signature. I'm happy thus far with the luminance image which I registered and combined using PI. I also performed a DBE, Histogram Stretch and a light HDR. The combine process was Average where in other softwares I usually use either Sigma Combine or Median Combine. Long story short, I've made combined masters of each red, blue and green channels using PI. My attempts to get a good color combine and proper color balance has been less than good. Maybe it's the data but I think I've better results using other software that I'm familiar with. Anyway, I've created and uploaded a zipped folder of these 4 images for anyone who cares to give it a go. The link for the download is http://www.astral-imaging.com/M95_29_Lum_SigmaReject_DBE_HDR1.zip The file size is 16.7mbs.

Thanks,

[sleshin]

Hi Steve,

I downloaded your R,G, and B masters into PI and created a RGB image. The color balance immediately after combining is off(magenta dominated), but after processing the colors look good/proper. I've tried to attach a jpeg of the processed result. Not meant to be a finished image, just to show that PI does a great job getting color balance right. The critical steps are DBE and Color Calibration. After these the color is mostly corrected. Then one can fiddle with color saturation boosts etc.

The steps I followed are:

1. Auto Histogram applied to each color master using the default settings. I believe this sets the black point for each image.
2. Channel combine to create the RGB master Image.
3. Apply the STF to the RGB image. This allows you to see the image and still keeps the data linear which is necessary for the initial processing steps. The color balance at this point looks wrong but that will get fixed.
4. Use Dynamic Crop to get rid of the bad looking edges that result following image registration.
5. Apply DBE. Juan's video tutorial on DBE available on the PI web site along with Harry's video are great guides to using this tool.
6. Apply Color Calibration. This is the critical step that will correct the color balance. Harry has a great video tutorial for this step. You can also do Background Neutralization at this point though I didn't for this exercise.
7. Apply SCNR to remove any excess green,
8. At this point can reset the STF and then apply a non linear stretch with the Histogram Transformation.
9. I used Curves to apply a slight color saturation boost.

I'm still a beginner with PI so I don't claim this is the best or only way to process this image. Just wanted to show you how you can get more correct color. I'm finding that the G2V derived color ratios used to create RGB images in programs like CCDStack or Maxim don't seem to be needed in PI. In PI, I combine the R, G, and B images 1:1:1 and then get more correct color balance with subsequent processing steps. Hopefully others will chime in to give you more suggestions.

Steve

[Harry page]

Hi

Here is my quick go

I just combined the rgb  Straight out of the box and did a Background neut at this point  

I think there is more in your lum as HDR wavlets needs appling better, you need to fiddle with the deringing to get better control.

Regards Harry

2 versions 2 different colours 

[sreilly]

Harry,

I noticed the color differences in my many attempts and wondered if it makes a difference in the galaxy sample being only galaxy without any background. The HDR is something I need to experiment with but at the same time should note that I prefer a less than over cooked look. I prefer lighter backgrounds as well. I did notice that if I wanted to do anything in PS to a PI image, it had better be saved as 16 bit first. That said, the PI 32 bit image looks very washed out in PS. Can't say why. The 16 bit looks the same in both. Of course you would only go to 16 bit when you are done processing in PI and ready to post to web.

[Nocturnal]

Hi,

you're better of doing DBE on linear data as you're dealing with removing an additive signal. Paraphrasing from one of Juan's videos here. My typical sequence is:

- auto STF
- crop
- DBE
- background neutralization (really not needed after DBE)
- color calibration
- histogram stretch
- disable STF
- small scale ACDNR
- conservative stretch to regain some range in the dark areas
- HDRWT if appropriate
- extract Luminance
- histogram on L frame to darken background, brighten target
- apply L as mask
- saturation curve to boost color
- SCNR as needed
- ICCProfileTransformation to set to sRGB

Of course I sometimes go back and repeat some steps but that's basically it. Simple stuff. When I save a copy of my image as 16b TIFF for publishing it shows up identical in PSP. If you don't get the same in PS then there may be a color management issue. Clearly a fully processed image should look virtually identical in 32, 16 or 8 bit uncompressed and look the same in different applications. For processing you want to stay in 32b as long as possible, naturally. But once stretched and processed 16b will do fine.

[sreilly]

Thanks for the detailed order. I'm guessing that this is from a one shot color camera? I would ahve thought that HDR would be best applied to the luminance only.Not sure what you are doing "apply L as mask". This is in PI? I'm familiar with ICC profiles but you have one for your camera or is this something that PI extracts in from color calibration?

[Nocturnal]

Yes, for an OSC but there's no difference AFAIK with your image except that you combine it yourself before you get started. By the time you get to step 1 you have a 32b float RGB image. Yes, I run HDRWT against L only but that's not strictly needed so I didn't specify it, just like I left the number of layers undefined. For me 2, 3 or 4 with 1 iteration seems best. Use the Mask menu to apply a mask. Or drag the mask image onto the tab of the image to be masked. I'm sure it's in one of the videos somewhere. In this context ICC Profiles have nothing to do with the image source. It just specifies how color is defined in your image so it can be accurately represented on screen or printer. As I understand it a freshly stacked 32b image doesn't need an ICC Profile because it's linear. Once you start squeezing that 32b range into 8b you need to specify a color profile to do this repeatably. Someone else will correct me if I got this wrong

[Juan Conejero]

Steve,

I did notice that if I wanted to do anything in PS to a PI image, it had better be saved as 16 bit first. That said, the PI 32 bit image looks very washed out in PS. Can't say why.

This is due to PS's inability to handle normalized floating point images generated by PixInsight. PS doesn't use the same data range that PixInsight uses internally to represent floating point pixel data. I don't know if you can configure the numeric range used by floating point images in PS. In PI you can, naturally.