Experiments with star rendering

[chris]

Below is an experimental replacement for the current 'scaled discs' mode of star rendering. The stars are rendered as gaussian discs with areas proportional to their brightness. Bright stars also exhibit glare.

This goes out to all the Southern Hemisphere forum members:


This mode of rendering produces the sort of images that you'd get with a CCD.

--Chris


[/img]

[t00fri]

Chris,

I think this looks very good, but I still have a question:

I magnified one of your stars



Now the question is this: By what requirement did you fix the gaussian width of the light distribution? Whatever the width is chosen, you may always assign the visible area over the Gaussian to the star's brightness. In other words why did you choose the "fringe" area so comparatively small, such that the impression of an almost sharp disk results on my 1600x1200 screen

Bye Fridger

[chris]

Chris,
Now the question is this: By what requirement did you fix the gaussian width of the light distribution? Whatever the width is chosen, you may always assign the visible area over the Gaussian to the star's brightness. In other words why did you choose the "fringe" area so comparatively small, such that the impression of an almost sharp disk results on my 1600x1200 screen

For bright stars, the central part of the disc is saturated. The height of the Gaussian is clipped to some maximum level, resulting in a wide saturated region surrounded by a relatively small unsaturated fringe region.

--Chris

[t00fri]

Chris,
Now the question is this: By what requirement did you fix the gaussian width of the light distribution? Whatever the width is chosen, you may always assign the visible area over the Gaussian to the star's brightness. In other words why did you choose the "fringe" area so comparatively small, such that the impression of an almost sharp disk results on my 1600x1200 screen

For bright stars, the central part of the disc is saturated. The height of the Gaussian is clipped to some maximum level, resulting in a wide saturated region surrounded by a relatively small unsaturated fringe region.

--Chris

Of course, there is clipping in your above star profile. But given that a Gaussian is determined by 2 parameters, its central height, and its width, I am trying to find out whether both are fixed by physical facts or whether perhaps one is fixed by "fiat" . Put in another way: you might for the sake of the argument, normalize the brightest rendered star to the maximal intensity the monitor can display. Then the width can be fit to provide a /visible/ area proportional to the star's brightness. Again one might discuss the brightness in logarithmic units or linear, etc...

Just trying to "knock" at the hidden degrees of freedom in your approach

Bye Fridger

[wcomer]

Fridger that's an interesting approach. For the same lens system shouldn't all background stars have the same half width, saturation notwithstanding? So one might let the halfwidth be fixed relative to the FOV; under the assumption that higher FOV is equivalent to higher resolution, and thus smaller halfwidths (though constant when measured in pixels.)

-Walton

[chris]

Just trying to "knock" at the hidden degrees of freedom in your approach

Indeed, there are some tunable parameters that need further consideration. There's the FWHM, which is set to 6 pixels in these images. There's also the saturation magnitude. Stars below this value have centrial peak brightness values less than the maximum representable pixel value. The meaning of the limiting magnitude parameters is obvious; how to choose a saturation magnitude isn't clear; for the moment, I'm just using the same calculation as in 1.4.0.

Fridger that's an interesting approach. For the same lens system shouldn't all background stars have the same half width, saturation notwithstanding? So one might let the halfwidth be fixed relative to the FOV; under the assumption that higher FOV is equivalent to higher resolution, and thus smaller halfwidths (though constant when measured in pixels.)

For the moment, I'm assuming that the halfwidth is a constant size in pixels, not a constant angular size. So, smaller FOV implies higher resolution.

Bright stars fill out the outer fringes of the Gaussian PSF to the point where they become visible above the noise level (here, that's just 1/255, the faintest non-black pixel value.)

--Chris

[t00fri]

Just trying to "knock" at the hidden degrees of freedom in your approach

Indeed, there are some tunable parameters that need further consideration. There's the FWHM, which is set to 6 pixels in these images. There's also the saturation magnitude. Stars below this value have centrial peak brightness values less than the maximum representable pixel value. The meaning of the limiting magnitude parameters is obvious; how to choose a saturation magnitude isn't clear; for the moment, I'm just using the same calculation as in 1.4.0.

Fridger that's an interesting approach. For the same lens system shouldn't all background stars have the same half width, saturation notwithstanding? So one might let the halfwidth be fixed relative to the FOV; under the assumption that higher FOV is equivalent to higher resolution, and thus smaller halfwidths (though constant when measured in pixels.)

For the moment, I'm assuming that the halfwidth is a constant size in pixels, not a constant angular size. So, smaller FOV implies higher resolution.

Bright stars fill out the outer fringes of the Gaussian PSF to the point where they become visible above the noise level (here, that's just 1/255, the faintest non-black pixel value.)

--Chris

Aha! That's about what I figured...

So there are still some things to discuss, when you are back...

I guess you noticed what I am having in mind instead of clipping: to normalize the central hight of the distribution to a physically prescribed (sensible) value and --rather than assuming a fixed fringe size-- determine the width and thus the fringe size implicitely by the requirement of the visible area being ~ brightness ( log10 (luminosity)?)

The resulting profile will be much fuzzier in general.

Bye Fridger

[t00fri]

Fridger that's an interesting approach. For the same lens system shouldn't all background stars have the same half width, saturation notwithstanding? So one might let the halfwidth be fixed relative to the FOV; under the assumption that higher FOV is equivalent to higher resolution, and thus smaller halfwidths (though constant when measured in pixels.)

-Walton

Yeah, eventually at very small FOV, we might want to see the Airy disks of the stars (including the diffraction rings) ...

Bye Fridger

[Malenfant]

Yeah, eventually at very small FOV, we might want to see the Airy disks of the stars (including the diffraction rings) ...

Bye Fridger

Why? It's bad enough that we're flipping back and forth between simulating the human eye and CCDs, now you're suggesting that it might be an idea to put in diffraction effects too?!

I think this 'observer effect' really needs to be fixed once and for all - at this rate Celestia is going to end up being a hodgepodge of different types of observer that would serve only to confuse the viewer. Either simulate the human eye in all cases, or simulate a CCD in all cases, or come up with some idealised observer - but let's not mix and match here.

[t00fri]

Yeah, eventually at very small FOV, we might want to see the Airy disks of the stars (including the diffraction rings) ...

Bye Fridger

Why? It's bad enough that we're flipping back and forth between simulating the human eye and CCDs, now you're suggesting that it might be an idea to put in diffraction effects too?!

I think this 'observer effect' really needs to be fixed once and for all - at this rate Celestia is going to end up being a hodgepodge of different types of observer that would serve only to confuse the viewer. Either simulate the human eye in all cases, or simulate a CCD in all cases, or come up with some idealised observer - but let's not mix and match here.

Malenfant,

I think you are misinterpreting our efforts here. Chris, myself, wcomer etc would certainly be discontent if things were to /remain/ in a fuzzy stage as they are admittedly now. Our discussions about these apparently tricky matters serve only the purpose of arriving at a clear physically sensible picture in the end.

The first attempt would clearly consist in sorting out what amount of arbitrariness might be hidden within a particular approach of the star intensity distribution. Next one can try to fix that arbitrariness by some kind of sensible physical criterion. Such a criterion could come from matching to the known response function of a CCD, for example.

My remark about Airy disks above was more of a joke, really

Bye Fridger

[t00fri]

...
I think this 'observer effect' really needs to be fixed once and for all - at this rate Celestia is going to end up being a hodgepodge of different types of observer that would serve only to confuse the viewer. Either simulate the human eye in all cases, or simulate a CCD in all cases, or come up with some idealised observer - but let's not mix and match here.

Honestly, the idea of /switching/ between different (standardized) sensitivity profiles appears as the only viable one to me, when a modelling of a /huge/ range of light source intensities in the Universe is the aim.

There seems to be nothing confusing to me in that people might be offered the possibility of switching from the eye to more sensitive detectors of light in a controlled manner. The only worry I might have here is the different spectral sensitivity that e.g. the unaided eye and a CCD are known to have.

As soon as we attempt a Celestia scenario that is able to display far away galaxies as dim as 18m along with glaring suns of -21m, on a "lazy" monitor, we /got/ to invent something pretty clever...


Bye Fridger

[ANDREA]

I'm an amateur astronomer and for this reason I'm following with interest your approach to star image definition.
IMHO, supported by years of visual and CCD approach to star imaging, I think that up today the stars shown in Celestia are absolutely unlike both types of viewing, i.e. visual or CCD.
The Guide8 commercial software is much closer to the real star appearance.
It gives three different commands to choose among:
1- star color saturation,
2- max and min star size,
3- mag range.
The max range adjustment allows the fine tuning of stars with slightly different mag.
E.g., if I choose to show stars up to mag 8, and I give a max range of 8, I'll see differently shown stars only if mags difference is bigger than 1.0, but if I give range 16, the shown difference will be 0.5 mag, and so on.
BTW, Chris' new approach remembers me the Akira Fuji's CCD images elaborations, where the luminosity of the luminous stars is fictitiously increased more of the less luminous ones, in order to better show the constellations shape, as you can see here (sorry, they are copyrighted, so I cannot load here):

Code: Select all

http://www.davidmalin.com/fujii/fujii_index.html

Sincerely I don't like it, because it's not real.
Equally the Southern Cross shown in Chris' image is not visible or imaged in such a way, IMHO, but nevertheless it's much closer to real than the actual rendering in Celestia.
My little cent.
Bye

Andrea

[wcomer]

Fridger,

If I've understood your proposal correctly then the following scenario would happen. If you are centered 2 ly form the brightest star, and then pull back to 4ly. The star itself would maintain a constant profile (assuming no other star became the new brightest star) despite loosing 75% of its apparent brightness. Furthermore it would continue to maintain this profile until such a time as some other star became brighter. Likewise, the profiles of nearby stars would be distorted under motion. For stationary views, this doesn't matter, but under motion, I have doubts about the aesthetics of normalizing against brightest apparent magnitude.

Also two bright stars near each other could cause some unusual scenarios. Let an observer be at (-3,0), star A be at (3,0) and star B be at (0,1); with star A six times as luminous as B. Let the observer follow a straight line towards A. Initially, A grows brighter, but then at (-1.6,0) B becomes the brightest star, after which the display of A will paradoxically grow fainter until (-0.3,0).

I apologize if I've misinterpreted the normalization scheme.

-Walton

[t00fri]

Fridger,

If I've understood your proposal correctly then the following scenario would happen. If you are centered 2 ly form the brightest star, and then pull back to 4ly. The star itself would maintain a constant profile (assuming no other star became the new brightest star) despite loosing 75% of its apparent brightness. Furthermore it would continue to maintain this profile until such a time as some other star became brighter. Likewise, the profiles of nearby stars would be distorted under motion. For stationary views, this doesn't matter, but under motion, I have doubts about the aesthetics of normalizing against brightest apparent magnitude.

Also two bright stars near each other could cause some unusual scenarios. Let an observer be at (-3,0), star A be at (3,0) and star B be at (0,1); with star A six times as luminous as B. Let the observer follow a straight line towards A. Initially, A grows brighter, but then at (-1.6,0) B becomes the brightest star, after which the display of A will paradoxically grow fainter until (-0.3,0).

I apologize if I've misinterpreted the normalization scheme.

-Walton

Hi Walton,

thanks for pointing this out. No that was certainly not the idea. Both yesterday night and as well tonight I felt "smashed", since I was digging myself for many hours each day through altogether 150 (!) postdoc applications...

So I hope during the weekend, I'll manage to express myself in a clearer way. Actually, it's quite tedious to discuss these issues at a more concrete level, without having a platform supporting mathematical notation in some natural way. Probably I shall retry tomorrow using my Maple worksheet environment instead...

Cheers,
Fridger

PS: Clearly in a scenario that wants to avoid oversaturation by design, one should not assume the area ~ brightness but rather substitute the area by the integral over the Gaussian intensity profile from the central point to that distance, where the intensity stops being visible. This gives the familiar erf function, of course...

[t00fri]

Hi,

perhaps in this discussion it's good to compare directly realistic photographic star images from extremely high resolution imaging with a telescope (Hubble) that is big enough in diameter in order NOT to increase the star disk sizes through diffraction and that does not suffer from atmospheric effects (<=> "seeing disc")!

Here is an example from the highest resolution Hubble photo of M 51 in undistorted TIF format (215MB (!), 11477x7965 pix ), also magnified 5x.


in comparison with Chris' saturated star disks above:


I guess it is apparent that in the hubble photo the stars are much more like Gaussians (rather than saturated disks) and the fringe region is MUCH wider than in Chris' model. Precisely what I was aguing above from theoretical grounds.

Chris perhaps oriented himself on star images that one can typically see in images taken with /earthbound/ amateur telescopes of smallish diameter (Astrophysics ). The star disks are then artificially much larger, since a) the Airy disk has a diameter of several arc seconds and b) there is the "seeing disc" due to atmospheric effects which is >~ 1", tyically! The above Hubble imaging is NOT degraded by atmospheric effects, of course.

Bye Fridge

[Malenfant]

That hubble image is horribly mangled by image compressed too by the looks of it... is there an uncompression version of that anywhere to compare it better?

[t00fri]

That hubble image is horribly mangled by image compressed too by the looks of it... is there an uncompression version of that anywhere to compare it better?

OK, I replaced the previous JPEG by the result from the undistorted original 215 MB (!) 11477x7965 pix Hubble photo in TIFF format!

No major qualitative changes, I'd say.

Bye Fridger

[Malenfant]

ok, thanks, yeah there's not too much difference....

[John Van Vliet]

all this is cool but my preferance is to have the stars shown as points
I just happen to like the look of stars as points , it is clean and crisp

just my 2cents

[suwalski]

all this is cool but my preferance is to have the stars shown as points
I just happen to like the look of stars as points , it is clean and crisp

I don't think the option to choose between star styles will have changed, so don't fret!