North America, Pelican, Birds nest, NGC 6997, IC 5068

[MilwaukeeLion]

Was testing different camera settings last week on few objects. Below 15 subs - 120 sec - iso 3200 combined with 8 subs - 180 sec - iso 1600.  doz each flats, bias, darks. mod DSLR, 80mm refractor. 

Does this workflow make sense? 

1. Batch processed both in PI - combined two final images 

2. Separated RGB channels in PI - removed stars from each channel.

3. Recombined starless channels - stretched image.

4. Recombined stars only channels - combined with stretched starless image & binned.

5. Messed about in PS.

Stars not bloated compared to my standard processing.  Having hard time figuring out how should be saving images while working on them.  Any suggestions? Work with fits or XIFS or tiff, 32 bit, 16 bit, free floating, integer, so many options.

 

[wimvb]

Very nice result. I save my wip as a PixInsight project. But i use pi exvlusively.

[vlaiv]

I would say FITS as it is standard astro interchange format - so most software should handle it properly.

Bit depth depends on "stage" and camera used (format of capture). Initial subs should be 16bit (unless for some reason you opted for 8bit format - but then better be sure you know what you are doing), and every stage after that should be at least 32bit.

I say at least 32bit because most software does not support double precision, but in due time we will see more and more software starting to support it if CMOS sensors, and particularly low read noise CMOS sensors continue hitting the market - then it will be feasible to have thousands of subs and that many subs require higher precision than can be provided by 32bit float point.

32bit integer will actually store greater precision than 32bit float point, but not many software supports this type of format, and one needs to be careful when coding support for it - most operations will require usage of long integers (64 bits) and scale back results to 32bit for storage. Ok, this is maybe a bit more info than needed use 32bit float precision for everything except captured subs and you will be fine for the time being.

[MarsG76]

That looks awesome.. perfect...

 

[MilwaukeeLion]

9 minutes ago, vlaiv said:

I would say FITS as it is standard astro interchange format - so most software should handle it properly.

Bit depth depends on "stage" and camera used (format of capture). Initial subs should be 16bit (unless for some reason you opted for 8bit format - but then better be sure you know what you are doing), and every stage after that should be at least 32bit.

I say at least 32bit because most software does not support double precision, but in due time we will see more and more software starting to support it if CMOS sensors, and particularly low read noise CMOS sensors continue hitting the market - then it will be feasible to have thousands of subs and that many subs require higher precision than can be provided by 32bit float point.

32bit integer will actually store greater precision than 32bit float point, but not many software supports this type of format, and one needs to be careful when coding support for it - most operations will require usage of long integers (64 bits) and scale back results to 32bit for storage. Ok, this is maybe a bit more info than needed use 32bit float precision for everything except captured subs and you will be fine for the time being.

No this is good info because im bouncing images between 3 programs, (PI - Nebulocity - Straton star removal). Easy to recombine starless image with stars only or combine 2 final results using stars or non stellar align in Neb. All handle fits but result out of Stratton is reduced by half size. From say 35 mb fits going in to 17 mb fits coming out.  Am I losing valuable info using Straton as opposed to taking the long route and removing stars manually?

[vlaiv]

16 minutes ago, MilwaukeeLion said:

No this is good info because im bouncing images between 3 programs, (PI - Nebulocity - Straton star removal). Easy to recombine starless image with stars only or combine 2 final results using stars or non stellar align in Neb. All handle fits but result out of Stratton is reduced by half size. From say 35 mb fits going in to 17 mb fits coming out.  Am I losing valuable info using Straton as opposed to taking the long route and removing stars manually?

Depends where you are at in processing pipeline. I'm not familiar with Straton, but recently have had a look at StarNet++ - it requires already stretched images.

Point with 32bit precision is too keep all the data in required precision up to point of stretching. After you have stretched your data, then you can go with lower precision - in the end, image ends up as being 8bit per channel simply because displays work with that kind of precision (and human eye can distinguish about half of that about ten million as opposed to 21 million of three channels of 8bit data, and there are some other technicalities like smaller gamut of displays vs human vision, but that is separate topic).

If you use Straton on stretched image then you don't have to worry that much about 16 bit format, but if not, I would suggest using different approach:

- make a copy of the image you want to work with

- convert it to 16 bit yourself and keep one copy of it

- open it in Straton and do star removal save as "clear" 16 bit copy.

- Use two above 16 bit images to create star mask and use original 32bit image to "paste" over only 16 bit data where stars once were in starless image - leaving much of image without stars unaltered. Then proceed to process / stretch such image and then in the end paste stars over finished image.

This way you will keep most important bits in high precision. You have to be careful though and it will probably require some pixel math to match 16bit data back to 32bit data. Best way to do it is to set your 32bit data to 0-1 range. When you convert it to 16 bit format it should be in 0-65535 value range (unsigned 16 bit). Once you have two images from Straton - one 16bit regular that you kept a copy of and one 16bit result of star removal, load them back into processing software change to 32bit and divide with 65535 to return to 0-1 range. Subtract two images and you should be left with star field only - use selection tool to select background, then invert selection to select only stars, then apply that selection to now 32bit version of starless image, do copy and paste over original 32bit version of image (copy of the first step).

Hope this makes sense?