Lagoon Trifid 32bit Processing

[avandonk]

When stacking with PixInsight the final stack can be saved as a 32 bit fit.

I did this and then produced a 32 bit RGB tiff with PI. This image was used to get rid of gradients and light pollution with DBE in PI. This is done in unstretched form. It seemed to be far easier to get a very good result with DBE.

Stretching was done in PI and saved as a 32 bit tiff. I then used StarTools HDR equalise and reveal thingies. The HDRreveal was used at a low setting. This was then saved as a 16bit tiff.

Photo Shop was used to apply HLVG. The image was then upsized by a factor of 1.5 in ImagesPlus. EasyHDR was used to do the final contrast curves etc. This is much better than PS.

I did not use any noise reduction at any stage after stacking.

Here is an eight bit result 14MB

http://d1355990.i49.quadrahosting.co...T_32b_Pr_M.jpg


Here is an animated gif showing the difference with DBE at 16 bit and 32 bit. 2MB

http://d1355990.i49.quadrahosting.co...14_04/LT32.gif




Bert

[alpal]

Much better job at stretching - well done.

[Peter.M]

Would have thought that if the camera ADC is only 16bits any processing software output cant create any more depth?

[alpal]

Originally Posted by Peter.M

Would have thought that if the camera ADC is only 16bits any processing software output cant create any more depth?

But the stack has 32 bits if you save it as a 32 bit FITS file.
I then use FITS Liberator to stretch my stack data.
Faint galaxy features can start to pop out.

[MrB (Simon)]

This is going to turn into a very interesting discussion!
I wouldn't have thought you could get any more than 16bits either, other than upsampled data from interpolation, which wouldn't be true data?
I'm a layman in this regard, looking forward to the opinions of the pro's.

[coldlegs (Stephen)]

Sorry Bert. Can't see the improvement. The original had black with a slight green tinge but know it's got a very strong blue tinge. You seemed to have dropped down the green and magenta and substituted blue as as well as darkened the image losing some of the cloud detail. Also seemed to lose a lot of detail in the brightest areas. Me, I'd probably tone down the red/magenta as it seems way too bright for my old eyes but that's me. Damn nice data you've got there.
Have fun with it.
Cheers
Stephen

[rcheshire (Rowland)]

No expert, but it is a nice image of an area with lots of variation.

Found this an intersting discussion

http://www.fxguide.com/forums/showth...some-questions

"the bonus of float is that you can have a value of 50, which would be 50 times as bright as white, in a similar way to the sun being brighter than anything else, yet will only ever be displayed as white in a daylight photo. The advantage here is you can move the brightness of the image around without killing the overall light values, so you change the exposure in a way similar to a camera"

[RickS (Rick)]

When you stack multiple subs from a 16-bit camera the result needs more than 16 bits to represent it accurately. Just thinking about the simple case of stacking two subs, you've gone from 65536 potential values in a pixel to twice as many (doesn't matter if you add or average).

Calibration arithmetic also results in non-integer results that can be represented more accurately by 32-bit floating point data.

Cheers,
Rick.

[avandonk]

Yes 32 bit floating point is the important thing. It gives a far greater dynamic range and finer numerical values i.e. levels. PixInsight gives the option when integrating or stacking to produce a 64 bit floating point image.

All the faint detail is in the bottom 1 or 2 percent of the full dynamic range of the data. The gradients and light pollution are there as well. So the more accurately you can evaluate the background gradients and light pollution the better it can be removed.

Below is a screen capture. showing the before and after image of two DBE operations.

The two small images are what was subtracted. All were stretched to make them visible. It is obvious how much light pollution I have in my backyard sky.

Here is a large version of the 2xDBE image 14MB

http://d1355990.i49.quadrahosting.co.../LT_2xDBE_.jpg

Bert

[avandonk]

If anyone is interested here is a 16bit version of the data. 58MB


http://d1355990.i49.quadrahosting.co..._2xDBE_16b.zip


The data has only had the gradients and light pollution removed at the 32 bit FP level.

I am sure that many others can do better than I have at producing a final image.

Bert

[avandonk]

With a bit more practice a better version 17MB


http://d1355990.i49.quadrahosting.co...4/LT_NM_L_.jpg



Bert

[strongmanmike (Michael)]

Fiiinally Bert, you are getting pretty good at this crazy art of processing (meant with great affection of course ) always knew you had it in there....

Mike

[avandonk]

If anyone else wants to have a go here is the 32 bit floating point tiff 480MB

http://d1355990.i49.quadrahosting.com.au/2014_04/LT_L_RGB_.zip


Bert

[avandonk]

Originally Posted by strongmanmike

Fiiinally Bert, you are getting pretty good at this crazy art of processing (meant with great affection of course ) always knew you had it in there....

Mike

Sorry Mike have no time to answer I have old data to do over again! This data is like girlfriends when you are a teenager. They still wriggle free while you are trying to work out what to do.

Bert

[Garbz (Chris)]

Very nice image, DBE has really done wonders with the higher bit depth.

Originally Posted by MrB

This is going to turn into a very interesting discussion!
I wouldn't have thought you could get any more than 16bits either, other than upsampled data from interpolation, which wouldn't be true data?
I'm a layman in this regard, looking forward to the opinions of the pro's.

Here's a simple though experiment. You have a 16 bit camera, and you take 2 photos and stack them to get the average.
One of the pixels comes up with the value 12580 on image one.
The same pixel comes up with the value of 12579 on the image two.

What's the average? In a 16bit image the average is 12579 which is clearly wrong. If you multiply the two number by 65536 first (interpolate) and then average you'll come to the correct value as you now have 65536 different shades between the original two numbers to help the maths work out.

What all this means is that Bert has a nice picture

[scopemankit (Chris)]

What telescope and exposures did you use?

[avandonk]

Originally Posted by scopemankit

What telescope and exposures did you use?

Astrograph is an Officina Stellare RH200 which has a focal length of 600mm and is F3. Clear aperture is 200mm.
FLI Atlas Focuser.
FLI ten position filter wheel CFW-3-10 with 50mm square filters.
Astrodon E series LRGB and HA, NII, SII and OIII 3nm NB filters. Also a continuum filter 5nm.
Camera is a FLI PL16803 which has a sensor size 36.8 X 36.8 mm.
The FoV of this system is 3.5 X 3.5 degrees.
Mount is a Software Bisque PMX.


Image of the naked system below.



Bert

[strongmanmike (Michael)]

Originally Posted by avandonk

Image of the naked system below.

Naked alright ..I didn't recognise her without her Burka on