Quote:
Originally Posted by marc4darkskies
Wow Ivo, it's hard to know where to start.
Not arguing that at all. It's clear that you've selectively reprocessed a portion of Bert's image - you've changed what you originally posted and moved the goal posts. I did not critique this image, I critiqued what you posted originally as a whole. So why make the point that sharpening is a good thing? Of course it is!
|
First point of your critique (which
is welcomed as long as it is based on factual information and
not opinion disguised as factual information) was that the stellar profiles indicated, in your opinion, the image did not require deconvolution. This opinion is demonstrably false, as I have hopefully demonstrated by the decon before and after.
Are you still maintaining the image as a whole does not stand anything to gain by deconvolution?
Secondly, of all people, you should know that sharpening and deconvolution are not the same things, just as 'cars' and Holden are not the same thing.
Different sharpening algorithms are used in different situations. In this case deconvolution was applied with a 2.7 Gaussian PSF in *all* of the images I posted (no changing goal posts here).
Sharpness and visual acuity are subjective measures. Many different factors govern sharpness and visual acquity, with many of them psychco visual. Deconvolution is an algorithm that helps reversing blurs (gross oversimplification) caused by seeing or the optical train, making sure that the viewer gets information that has optimal retinal focus to begin with.
Then there are algorithms that use psychovisual tricks to enhance visual acuity, by manipulating local contrast (Retinex, local histogram manipulation) or enhancing feature sizes on a global scale (wavelet manipulation).
Quote:
I see that you've fixed it. But wait! ... See the first 3 star images below - from near the top of the frame. The first is my quick & dirty, the second is your first attempt (the one I critiqued) and the third is your "fixed" version. The one that I critiqued (#2 image below) clearly shows less faint stars and less nebulosity - you have hidden or erased information from Berts data.
|
This is false. Compared to the linear and calibrated version of the data, no data was destroyed and data was still brightened.
More so, your version shows 'fatter' stars than needed (a feature I have noticed in more of your images) and less optimal use of dynamic range with no data using the lower 8% of the available dynamic range.
Quote:
Your second version (#3 image below) shows more neb but the stars are now a wonky shape. To me it looks like your reprocessing is more than just a star mask(?) #3 still has less stars and less neb than it should (IMO).
|
It's a different data set that Bert posted - process it yourself and you should see the same 'wonky' stars!
Quote:
But I didn't say that Ivo. Decon and wavelet do not automatically overcook an image. I meant that you had applied (IMO) too much sharpening to the image, period. Like I said, selective and controlled application of any sharpening technique is important - I do it all the time. Would I sharpen this image if I was processing it? Definitely.
|
My bad... it was the word 'consequently' that lead me to believe you thought one ('overcooking') was a consequence of the other ('using wavelets on top of deconvolved data'). I thought you were trying to state a fact rather than an opinion.
You could have fooled me though when it comes to judicious use of sharpening of some of your own images. It is clear that on smaller scales you indeed use some sort of sharpening, but on larger scales your images are disappointingly lacking detail, especially for the integration times quoted. Your Helix being an extreme example (though i appreciate that was recorded in 2009), but even your recent M8 is indistinguishable from a more amateur setup and much lower integration time which directly has to do with your processing;
If all you are doing is a global stretch with small bits of small scale sharpening, then beyond a point of fidelity, you better data is (both visually and mathematically) just wasted. It is only when you start resolving the dynamic range challenged areas in your image (such as M8's core), the fidelity of your data is being used.
The latter paragraph is fact, not opinion. Happy to show the math behind it.
Quote:
Look at the 3 neb images below. The first again is my quick and dirty (no decon or wavelet). The second (#5) is your first version (and the one I critiqued). You see the halos and you immediately notice the huge amount of additional detail. Is it all real, possibly, but I doubt it.
|
Then it is really time to get with the program Marcus. It is real.
Here are similar results for a program that is not StarTools (PixInsight);
http://astrofoto.euweb.cz/pixinsight...omparison.html
StarTools' algorithm is sublty different, but uses the same sort of Local Histogram manipulation at its core.
Quote:
Have you corrupted the tonal balance? Yes, IMO. Eg: The blacks are lighter.
|
Changed the tonal balance? Yes - ofcourse! As I indicated above, you'd be doing yourself and your data disservice if you didn't!
Quote:
In your second version (#6) the stars are still a funny shape and the blacks are now almost gone. You're flattening the dynamic range too much IMO. #6 also looks like the fine structure is less sharpened than #5 so I doubt this is just the application of a different star mask.
|
Yes, you're correct - it's a different data set and I used slightly different settings as a result.
Quote:
Unfortunately, in your second version you've exaggerated the large scale structure too much for me.
|
I think you mean I exaggerated small scale structures too much in lieu of larger scale structures. I think you were just trying to say that your taste in aesthetics is more about showing detail in its larger context in lieu of less detail? Your images certainly seem to do this?
Even so, be careful when taking a crop when critiquing scale manipulation is like critiquing a classical concert over a phone line; you need to at least be able to hear (see) the full frequency range over which the work is performed and affected (which in this case slightly larger than your crop
).
Quote:
How would I define overcooking? One or more of the following: Oversharpened, over flattened dynamic range, loss of information, creation of structure that isn't real (even if just tonal in nature).
|
Then we're on the same page it seems.
Quote:
Nice spruik Ivo, but I'm not critquing your software product - I'm not qualified. I critiquing your processing - that's all. Not something I'll soon repeat I might add.
|
StarTools is a (consciously) not-for profit loss-making endeavor to enthuse more people for our hobby (and offer people stuck in the Photoshop era a very-low-cost way out). There's no commercial motive, so I don't need to spruik for commercial reasons.
Now that that is out of the way, re-read it as an explanation for why the processed image looks the way it looks.
Quote:
Ooops, you just lost me.
|
Don't state your opinion as fact and we're all good
Bert has graciously posted this data for all of us to learn from. I don't want anyone to feel any of their attempts are inferior, wrong or 'unnatural', unless demonstrated through fact, not opinion stated as fact or false authority. Even more so when we're talking about narrow band data that is outside the visual spectrum and that no human eye has ever seen. Any references or claims pertaining to what is or isn't a 'natural' look for such data should be taken with the largest possible grains of salt...
You don't like the look of my (quick) processing - that's cool. Feel free to tell me this - I feel the same about yours. Let's keep things factual and informative though for those who may not have gotten to grips with narrow band data, dynamic range optimisation, wavelets, deconvolution, etc.
Cheers,