HDRCombination Ignoring Faint Detail from Long Exposure Image

[mcintyre_sj]

I have two stacked images of M31 from DLSR data that i am trying to combine using the HDRCombintation process (not the older script utility). But HDR seems to be ignoring the faint detail from the longer exposure.

One image is a stack of 3hrs at ISO1600 with 6min subs and the other is 4hrs at ISO800 also with 6min subs. The ISO1600 shows move faint detail in the outer arms - although not all that much more. The ISO800 image shows much better detail in the core.

I have calibrated and stacked each image set (using ImagesPlus) and created a single tiff file for each image set. The ISO1600 image has a much higher average pixel value and i have not [yet] attempted to normallize the two images before combination. I then tied combining the two images using HDR.

HRD does correctly select the unsaturated areas of the core and central region from the ISO800 image. But it seems to ingore the faint data from the ISO1600. I have visually examined the three images - 2 original and HDR combination - and the HDR version is nearly the same as the ISO800. To further test, i subtracted the ISO800 from the HDR image and the result was alsmost black. Doing ISO1600 - HDR however shows a lot of stuff left over, meaning the HDR image did not pick up a lot of data from the ISO1600 image.

More fustrating, noise in general does not seem to be improved with the HDR image looking almost exactly like the ISO800 image. I am perplexed that doubling the data has not increased the SNR - ie made a smoother image with more detail. 

I have tried bumping the binarizatoin threshold from .8 to .9, with not decernable difference in the faint stuff. I think it might have selected less of the unstaturated data though - which is not what i wanted. I have not tried fiddling with any other parms.

Suggestions?

[Juan Conejero]

Hi,

To provide a really useful answer I'd need to take a look at the two images you're trying to combine. Could you please upload them in raw form (or in the format and conditions you're using them with HDRComposition)?

I may be wrong, but this case looks perhaps similar to a previous one, at least from what you're describing:

http://pixinsight.com/forum/index.php?topic=2466.0

Besides the applicability of HDR composition to your images, I detect a conceptual error in your post:

noise in general does not seem to be improved with the HDR image looking almost exactly like the ISO800 image. I am perplexed that doubling the data has not increased the SNR - ie made a smoother image with more detail.

HDR compositions, in general, don't improve the signal-to-noise ratio in the combined images with respect to the individual frames. When you perform an HDR composition of two or more images you are not adding or averaging data; you are actually replacing saturated pixels or nearly-saturated data with scaled nonsaturated data. Actually, it is not uncommon that the general SNR worsens with respect to the longer exposures, since after all, they are being replaced partially with shorter exposures. Dynamic range and signal-to-noise ratio are two unrelated concepts.

[mcintyre_sj]

Actually my "conceptual error" was more wishful thinking. You explanation of the difference between HDR and an sum or averaging process confirms what i thought was happening.

Perhaps i'll try some pix math to see if i can combine the images in some sort or averaging process that also deals with the saturated from the longer exposure image.

Where/how does one upload files? Is there a pixinsight forum upload area or do i use some general ftp upload site?

[Juan Conejero]

Hi Stephen,

Well, since we migrated to a new virtual private server this year we no longer have an anonymous ftp account; our new hosting provider (Inmotion Hosting) forbids it for security reasons.

It is obvious that this forum is needing an upload area available to all registered users. While I find the time to set it up, please use a public/free file upload server, such as MediaFire or EasyShare (just to name a couple known sites; we have no affiliation with them).

[zerro1]

Something to think about with that M31 data.

I had been messing around with this data for a month or so. In my various approaches including DSS, and HDR merge(photoshop). I tried layering in the individual stacks, but I had lots of trouble getting precise alignment.  So I tried an approach that someone suggested was "not" the way to go about it.
In PixInsight;
Having completed the caliration and debayering, I registered en-mass 47 sub frames that included 10 minute, 6 minute and 3 minute subs. Then integrated en-mass into a single image. Since the core was still blown, through a bit of struggle I was able to layer mask a stack of 10 second subs (in photoshop) to pull the core back in. In effect it's not a "by the book" HDR, but I think it came out quite well.
Scope= AT65EDQ | camera = modified canon 1000D

I going to see if I can get the same kind of result from some m31 Data I have.

Robert   

[Juan Conejero]

Hi Robert,

This is an ideal case for the HDRComposition tool in PixInsight. The basic procedure would be:

- Register all individual frames with StarAlignment.

- Integrate the 10 minute frames. Do the same with your 6 minute, 3 minute and 10 second frames. Now you have 4 "master" images covering a large effective dynamic range.

- Now feed HDRComposition with the four master images and you'll get a linear HDR image in a similar way to this example.

With the HDRComposition tool you get a linear HDR image, which is a very important difference with respect to other methods and tools. Another advantage is that the processes of HDR composition and image scaling are completely automatic and algorithmically defined, and hence free of the arbitrary manual manipulations that are characteristic of other methods based on layering nonlinear images in other applications.

[mcintyre_sj]

I eventually just normalized the two stacked images using ImagePlus normalize function for contrast and brightness. Then did a combine using a simple average. This produced the best overall image. The SNR is marginally better and some of the faint detail from the long exposure is showing up. The core is almost as good as in the shorter exposure with the clipped area of the core being much smaller than in the long exposure

[Juan Conejero]

did a combine using a simple average. This produced the best overall image

Then an HDR composition is useless with these data, as happened in the other case I mentioned above.

Note that you can perform the same task in PixInsight with the LinearFit and PixelMath tools. Just apply LinearFit selecting one of the images as reference (normally the longest exposure) and use a simple sum in PixelMath (Image1+Image2) with rescaling enabled. Ideally, this task should generate a 32-bit floating point result, even if you're averaging two 16-bit images.