Experiments with star rendering

[chris]

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.

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.

The stars in my images *are* Gaussians, clipped to some maximum CCD signal level (which is what I meant by saturated.) The saturation effect is obvious in the Hubble images because the fringes are colored, while the center of the star disc is white. This occurs because at the center of the star, the signal max is reached in all of the filtered images composited to create the final color image.

Here's a plot of the Gaussian (sigma=sqrt(1/2)) for a star right at the saturation magnitude and the resulting star disc:




Now, here's a Gaussian scaled for a star one magnitude brighter than the saturation magnitude. The FWHM for the Gaussian is exactly the same, but the star appears larger as the scaling causes the 'tails' of the Gaussian to be filled out (that is, there's a larger region of the Gaussian with a value greater than the minimum pixel value 1/255.)




And for a star two magnitudes brighter than the first:




Notice how in the last image, the fringe is small relative to the size of the saturated inner disc, just as it is for the stars in my Celestia test images.

Now, I think my reasoning is sound so far--do you disagree with anything up to this point?

I don't believe that the extended fringes of the star in the Hubble image are from the star's PSF. I that they're caused by secondary effects, such as internal reflections within the optical system. These effects are the same ones responsible for lens flares that appear in photographs of very bright objects. I've simulated this by adding an additional glare halo (again a Gaussian) for stars above a certain brightness. I think it looks reasonable, but I'm uncomfortable that I don't understand the effect better and that I've had to tune it to look right. I do however feel confident that there really is some secondary effect contributing to the appearance of stars in CCD images. If you disagree, I challenge you to show me a Gaussian that fits a cross section of the star disc in your Hubble image

I'd be happy to be proven wrong . . . Purely Gaussian star images are easier both theoretically and from an implementation standpoint.

--Chris

[forceflow1049]

I think that at some (perhaps user) designated distance from the viewer, the stars should revert to point or fuzzy point style. This would IMO be a good way of simulating the distance while conserving systen resources for those with lower end systems.

[Sirius]

Some time back, I experimented with some atmospheric effects in celestia - "star twinkling"; so to speak.
As Celestia already determines if and how deep the observer is inside an atmosphere, it was quite easy (~3 loc) to scale this "twinkling" effect.
It could also be used to superimpose other atmospheric effects to an out-of-atmosphere gaussian "picture"... and if one troubled enough to check the "local temperature" (on the planet) one could also do the "twinkling" right.

Besides that, what I'd like quite a lot would be a setting of "telescope diameter"; causing that anything that is smaller than a pixel to have a nice diffraction pattern. This would (on the educational front) clarify the necessity for bigger telescopes :-)