My Tutorial PixInsight Workflow for DSLR Images

[sctall]

HI

I am bit confused on the context I will use with DSLR.
Please bear with me here:

Following your HDR tutorial, In step 5(Creating the HDR composite), you made 2 HDR images. I assume with my DSLR, I would do this only once.

So following your original DSLR tutorial, I would do everything normal on both long and short exp images separately.    DBE, BN,and CC.
I would then create an HDR image using both images. Once I have that image,
Then starting with step 4 (Extracting a Luminance image from the RGB image), in the DSLR tutorial, create a luminance image and follow all steps as normal from there on.

Scott T.

[kayronjm]

HI

I am bit confused on the context I will use with DSLR.
Please bear with me here:

Following your HDR tutorial, In step 5(Creating the HDR composite), you made 2 HDR images. I assume with my DSLR, I would do this only once.

So following your original DSLR tutorial, I would do everything normal on both long and short exp images separately.    DBE, BN,and CC.
I would then create an HDR image using both images. Once I have that image,
Then starting with step 4 (Extracting a Luminance image from the RGB image), in the DSLR tutorial, create a luminance image and follow all steps as normal from there on.

Scott T.

Precisely correct. The idea is that you have to use two full, separate images (preferably after DBE and CC) to produce a HDR result. This HDR result can then be considered the image to process afterwards. Do make sure that after producing the HDR image, you stick to 64-bit all the way throughout as you will need to maintain the extremely large dynamic range provided by the 64-bit format.

Do note that once you perform the histogram stretch, it will appear as if you've lost the nebula core detail once again. Do not despair, this is recovered later with the use of HDRMultiscaleTransform (described in step 7 of the HDR tutorial for the Great Orion Nebula).

[sctall]

Thank You
That worked great
Awesome

Scott T.

[georg.viehoever]

Hi kayronjm,

some comments on your tutorial after a more careful look.

Step 1:
- "it is a good idea to capture dark and bias calibration images": Right, but you also need flats. They are mandatory.
- "Capturing a number of these (5 or 10) is a good idea to average them individually": You need more. Use 50 bias frames, and at least the same number of darks and flats as you have lights. Otherwise, the masterdark/flat will introduce most of the noise into your final result
- " this tutorial discusses how to take your target, dark and bias images and produce one final, calibrated result.". You should mention the BatchPreProcessing (BPP) script in PI here. Your tutorial uses DeepSkyStacker, which is fine, but PI will give you more control. When you mention BPP with DSLR, dont forget to check "CFA" along with RGGB debayer and uncheck "Scale Darks".
- "It is important that this final image be in 16-bit, and either TIFF or PNG is a good format. An 8-bit JPEG will be extremely, extremely difficult to post-process later due to the inherent issues presented by the image compression." Wrong. PNG is also an 8 bit format. Your stacking result should be TIFF or (in my opinion preferable) FITS 16 or 32 bits. The result generated by BPP is a 32 bit float FITS file.

Step 2:
- "STF AutoStretch": You should unlink the channels before doing this autostretch. This will remove the orange color of the background.
-"The light pollution captured is very evident": What you see is mostly vignetting (no flats), and probably the result of minor image motion compensated for by the registration (bands at borders).
- "Therefore, save the image as a new file and select FITS for format. Enter a new filename such as the original one with _Crop added on to the end. When asks for bit rate, select 16-bit unsigned integer": Before you do any further processing (in particular: noise reduction, HistogramTransform), you should convert to 32 bit float format. This better preserves values and dynamic range than 16 bit integer. Use Process/Image/SampleFormatConversion. Also save as 32 bit float.

Step 3:
-  "Change the Default sample radius to 15 and Samples per row to 25 (to add more sampling points, of larger radius": I am not sure if it is necessary to increase these values from their defaults. The only value that I usually need to change (with heavy light polution or imperfect flats) is Tolerance (1.5 or so).
- "they can either be moved around or selected (by clicking) and deleted individually". What I usually do is apply the current samples ("subtraction", "discard background model", but not "Replace target image"), check the result for artifacts (such as dark holes caused by samples placed on stars), adjust samples, repeat this until things look good. Only then I apply with "Replace target image" checked.

Step 4:
-"Extracting a Luminance image from the RGB image": I am not so sure about the value of extracting a luminance image from an RGB image. Maybe others can comment.

Step 5:
-"64-bit IEEE 754 floating point": Rarely necessary. 32 bit float is sufficient except for some HDR images .

Thanks for the nice tutorial!
Georg

[Harry page]

Hi
First of many thanks for your time and effort in doing this  8) , But I would make a few observations

1) As just mentioned , unlinking the channels of the image will give a better representation of the data and will make it easier to work with

2) When using DBE I always say the quality of samples is better than quantity , and perhaps a bit longer placing the samples avoiding nebula etc would reduce the number required and avoid possible errors ,

3) Always look at the DBE background to inspect for errors

4) When doing noise reduction ( at wavlets ) its more normal to use a lum mask as this is more easily controlled and adjusted

5) On using Background neutralisation you should select a bit of the image with no nebula or anything other than free background sky or you will not get correct results , a very small sample will work ( less than 15 pix square will work)

6) I am sorry I do not think your colour calibration method is valid PI CC works on the assumption that a) a large enough selection of stars will average out white or b) The total output of a near by galaxy is white the selection of a piece of nebula is not correct sorry

7) HDR wavlets works on looking at the whole image ( hence why previews do not work well ) so be careful about creating harsh masks with course transitions ( I use a lum mask with a star mask or just a star mask)

IMO  I agree unless the Lum image brings something new to the party ( i.e genuine lum) I see little benefit from separating it out for separate processing 
a lot of tools in pixinsight are capable of processing the lum part of a image while combined ( acdnr as a example) 

I would be interested in other peoples opinions

Kind Regards
Harry Page

[pfile]

i like processing the L of a DSLR image separately since i feel like i can handle the background noise a little easier that way. this may be because of my LP skies.

one important thing - as juan has pointed out though, L* is not L and by due to the default RGBWorkingSpace of an RGB image, extracting L* gives you something that has too much green weight. so you need to change your RGB weights to 1,1,1 before extracting L*, or simply integrate your RGB images together using the default noise weighting in ImageIntegration for best SNR results.

on the other hand Alejandro Tombolini has shown a great technique to remove the large scale color noise from an RGB image so that might be the right way to go with DSLR images with a lot of background noise.

rob

[Batch]

Rob, Do you have a link for Alejandro Tombolini's technique (or perhaps can you outline what it involves)? Thanks.

Bill

[pfile]

it's in here somewhere, i'll have to search for it. it might be in one of the tutorials that he aggregated in a single thread.

the gist of it is to mask the foreground and stars aggressively, leaving only the background well-exposed. then you use atrous wavelets to remove large scales with the target set to Chrominance.

rob

[pfile]

ah here is the sticky thread with all the tutorials

http://pixinsight.com/forum/index.php?topic=5283.0

rob

[Batch]

Thanks very much Rob.....there is a lot of interesting information there.

Bill

[sctall]

I have to attest, that kayronjm's tutorial made a huge difference in my final image.
I think it a great tutorial to start to learn with.
I did all the preprocessing with BPP, and then used all the steps in the tutorial.
As far as the Luminance, it made a huge difference. In both noise reduction and contrast.
And after doing the BN and CC following his steps, my balance was dead on.

I am sure that there are some tweaks and improvements to be made, and not all images will perform the same, but this was an excellent tutorial for me to use and follow.

Using advice from the experts, and understanding how to incorporate them into this tutorial can help out.

Scott T.

[kayronjm]

Thank you all for your observations and suggestions. I have taken them onboard with regards to the tutorial and made some changes to the text and therefore instructions. I do not precisely remember why I did not mention flats initially, but last night I remembered and even before reading this thread with all your replies, I did add it. I have emphasised flats further, though (I always use them myself). Throughout, I do not ask that the user click Unlink Channels before an STF AutoStretch simply because the AutoStretch is done to see where there's some nebulosity or stars to avoid in DBE. I appreciate that unlinked channels makes sense prior to DBE and ColourCalibration to give a more realistic look of the data, but as long as the user is able to see where to place DBE samples and where not to place them, it suffices in my view.

With regards to the 64-bit format, I do agree it is overkill in a lot of cases. 32-bit should indeed suffice in the most part. I do however not see a problem, per se, in using 64-bit over 32-bit, besides some extra HDD space occupied by the files. I have also made is clear in the tutorial, which was not previously made, that extracting a Luminance image from the One Shot Colour RGB image (e.g. from a DSLR) does not constitute a real Luminance image as in those captured by monochrome CCD cameras with a proper Luminance filter. The post-processing technique does however seem to work very well in terms to bringing out detail and removing noise, so it is something that the user CAN do if desired. I myself always work with a monochrome CCD camera and do most definitely see the value in a proper Luminance image, by all means!

Oh and thank you for pointing out the PNG format issue. I removed the suggestion entirely. 16-bit TIFF it is, for initial import into PixInsight. That or the beloved FITS I always use myself, of course.

Finally, I will look up Alejandro Tombolini's posted technique for removing large scale noise in RGB images. With due credit, of course, this may be a great addition to the tutorial. My thanks once again!

EDIT: After reviewing Alejandro Tambolini's technique for removing large scale noise in chrominance, I added it to my tutorial under the post-processing of the RGB colour image step (due credit given to Alejandro for the technique, of course - with link to his post).

[sctall]

Kayronjm,

Since you have a pretty good bead on this stuff, have you done anything with the TGVDenoise tool for “Noise Reduction.
I have been experimenting with it, but it is a hard one to get dialed in on.
The few times I have, I think it works really good.  For Linear though, you definitely need a support image. One more thing that the parameters are tough to get right.

Scott T.

[kayronjm]

Kayronjm,

Since you have a pretty good bead on this stuff, have you done anything with the TGVDenoise tool for “Noise Reduction.
I have been experimenting with it, but it is a hard one to get dialed in on.
The few times I have, I think it works really good.  For Linear though, you definitely need a support image. One more thing that the parameters are tough to get right.

Scott T.

Hi Scott,

This is one thing I practiced with a few days ago. I understand it, but I agree that tweaking parameters can be difficult, though you can get a great head-start by checking the background standard deviation via a Preview box encompassing background and using the Statistics tool on that Preview segment of your image. This helps define the value for "Edge protection". This video tutorial on the tool is useful to get a head-start:

http://www.ip4ap.com/IP$AP_Flash/Pixinsight_2/PI-13_Nonlinear-7/PI-13_Nonlinear-7_player.html

I do however remain to play around with it extensively and compare results with my usual denoising routine. I will write up a tutorial on it once I get this sorted, though I started work again yesterday and time will be a constraint for the next 13 weeks (I'm a secondary school teacher).

Kayron

[Gardner Gerry]

Kayron,

Thank you for sharing your tutorial. I spent a good part of my Saturday reprocessing 4 year old DSLR data of the Horsehead and Flame region following the tutorial step by step. I was a bit skeptical at first, but soon realized there was much more in my data than I ever dreamed. Now of course I have years of old images to reprocess!

Thanks again!
Gardner