Something a bit different:- Reprocessing a professional image

[sjastro]

The Carina dwarf galaxy is a notoriously faint object as described at the ESO website. The natural skyglow at Chile where the image was taken is fainter than the surface brightness of the galaxy.

http://www.eso.org/public/images/potw1126a/

I measured the skyglow in the image and subtracted it out. This allowed a more aggressive stretching of the image. The Carina dwarf while still faint is clearly more visible.

The left hand thumbnail is the untouched image, the right hand image is skyglow subtracted and stretched.

Higher resolution images are found at:-

http://users.westconnect.com.au/~sjastro/various/carinadwarf.jpg
http://users.westconnect.com.au/~sjastro/various/carinadwarfprocess.jpg (processed image)

I could have spent more time fixing the brighter star images.;)

Regards

Steven

[CraigS]

G'Day Steven;

The difference is huge !

How did you measure the sky glow from someone else's image ?

Cheers

[sjastro]

Hi Craig.

Welcome back. The sky glow of the image can measured as a background pixel count using a CCD image capturing program such as CCDops.
While this is not an absolute value with images that have already been processed, the variation in the background count gives me an idea of how much skyglow can be subtracted without effecting the overall image.

Regards

Steven

[renormalised]

It's a faint one, alright!!!. Nice processing:)

[CraigS]

That is very cool.

I'm by no means an expert in this kind of processing however, looking at the 'before' and 'after' images, you can see how the really, really faint stars emerge when the background is subtracted .. which would mean that the process doesn't introduce too much more noise … which might then lead to artifacts appearing in the final image.

They say the raw image is the best taken so far .. if that is the case, then I reckon Steven's trumps it !

Patent the method/algorithm, Steven … quickly !!

Cheers

[sjastro]

Here is some science. Equalizing the data so that the Carina Dwarf galaxy had similar stellar densities in both images and measuring the sky background characteristics produced some interesting results.

Carinadwarf.jpg: Sky brightness:- 5574 ADUs, Noise:- 1612 ADUs
Carinadwarfprocess.jpg: Sky Brightness:- 34 ADUs, Noise:- 4 ADUs.

Not surprising that the Carina Dwarf was lost in the skyglow for the unprocessed image.

Regards

Steven

[jase]

Wow! How diffused and faint is that! Lovely work Steven. The logic in subtracting the sky glow has certainly improved the scene. Awesome!

[sjastro]

Thanks Carl and Jase for the feedback.

Regards

Steven

[irwjager]

That is really, really cool Steven!
This made me wonder about two other techniques that could be used to bring out that hidden beast. They are probably somewhat less scientific and hinge on some assumptions, but I'd love to hear what you think.

A galaxy is far away. We can therefore assume that most, if not all, stars belonging to that galaxy are small. Because we know the population of small stars in the image is skewed in certain places in the image because of the presence of the hidden galaxy, we can bring out the galaxy by emphasizing small stars.

The latter can be accomplished by simply generating a star mask that only contains small stars. Having isolated only the small stars we can then manipulate them in any way we wish.

Trick #2 is 're-synthesising' (if you will) a super structure from the density distribution of the small stars. For this I used a trick I sometimes like to use to 'liven up' nebulas somewhat. I layer the image with a blurred copy of itself in screen mode and apply a luminance mask so the screening is only applied to the darker parts of the image. The Gaussian blur in this instance spreads the energy of each star over a larger area, creating a large scale super structure which is easier to see.

Higher res image at http://www.startools.org/download/potw1126a_ST.png

[sjastro]

Well done Ivo.:thumbsup:
I checked a region of the background in your higher resolution image and found Sky Brightness is 25.7 ADU and noise is 1.6 ADU.

The important point is you haven't accentuated the noise or the background.

I think you should seriously consider sending the image to ESO.
I checked their copyright conditions and manipulating their images is OK.

Regards

Steven

[irwjager]

Thanks Steven,

The difference between yours and mine is that yours would pass muster with the purists, whereas mine most likely won't.

The creation of a star mask with just the small stars is a somewhat subjective matter. The method, though highly automated, is not exact in its highlighting/masking of stars. The whole method described above is founded on principles of statistical probability and random distribution, but I cannot say for sure I am bringing to the fore only the galaxy's stars, nor can I guarantee I get all of them.
The method is good enough to approximate the galaxy's shape and star distribution. However, it is not a 100% accurate representation of reality. Then again, no image is... :P

[sjastro]

If you check the histogram of the ESO image you will find it's clipped, so the data wasn't 100% to start with. That's not surprising as processed professional images for public consumption aren't necessarily any better than what an amateur can accomplish.
I wouldn't sell your efforts short. Your image is able to differentiate the galaxy from the foreground stars which is the ultimate objective.

Regards

Steven

[strongmanmike]

Interesting work you two, thinking outside the square again Ivo ;)...I like both results :thumbsup:

Mike

[irwjager]

Steven,

I sent ESO a link to the image this morning and got a very nice reply just now from Lars Lindberg Christensen/ESO

The response is as I expected - the image as-is, is of little scientific value (and I must agree). However, I replied with a method that is similar and would allow for the specification of a confidence interval of the results, such as, for example, the popular 95% confidence interval often used in scientific research.

This method would concentrate on the density of distribution of small stars.

We would convolve just a single pixel per star, using a kernel radius that would result in a 5% energy per-pixel convolved result. We would then add this result to the image (e.g. 20 pixels in the immediate vicinity of the star would be brightened by 5%). As such, the 95% confidence interval requirement would be met (e.g at worst we would be adding 5% false signal to the image in case we erroneously selected a star/artefacts that wasn't supposed to be selected). Dense concentrations of small stars would cause a significant local brightening, causing a 'super structure' to emerge.
We can leave this super structure in the image, or we can leave out the super structure and measure the brightness of the super structure where small stars are located and brighten them accordingly.

Care must be taken to select only small stars. Care must also be taken to heal any parts of the image that are affected by big stars (and halos) in a way that does not upset local star density (e.g. these parts of the image need to be replaced by an approximation of small-star density), so that convolution of that area results in a neutral patch. Despite these 2 challenges I do believe this would be a viable approach.

Then again, I may have missed something (to be honest, probability theory isn't my strong point) and may just be full of it... :P

[DavidU]

Fascinating stuff guy's.

[sjastro]

Ivo,

It's excellent you got a reply, even though I don't agree with it....:shrug:

First of all with regards to scientific standards, the ESO image doesn't stand up either. As I mentioned previously their image is clipped, (resampling does give a "decent" histogram however). So they have lost quite a of bit detail in the galaxy.

Secondly on the question of artifacts, processing an already compressed processed JPEG image doesn't help.
A way of dealing with this issue and incorporating your own ideas is to process their raw FITS data.

I wonder if ESO would be obliging?

Regards

Steven

[Ross G]

Amazing photo Steven.

Your imaging is so extreme!

Thanks.

Ross.