[Author Prev][Author Next][Thread Prev][Thread Next][Author Index][Thread Index]

Vinetting and SGBNR info (Was: Re: [APML] Pleiades response to Processing Challenge: LMC)



Bert, Don,

We have an article on vignetting correction at our website, but, 
unfortunately, this article is available in Spanish only ---we haven't had 
time enough to translate it. However, we included a lot of figures and used 
the English version of Photoshop 6.0 to make them, so I think you won't 
have too much problem to understand the techniques described. The article 
is here:

http://www.pleiades-astrophoto.com/doc/vignetting/es.html

MS Internet Explorer >= 5.0 / Netscape >= 6 required. We'll be glad to 
answer any question or doubt you may have. In spite of language, I hope 
you'll find that article of some help.

Regarding SGBNR, we apologize for having no documentation specific for 
version 1.0.9 yet. We are finishing our new house now. Whoever has built 
his/her own house knows what a sort of nightmare this means. However, I 
have some free time now and the aim of making things clearer for you and 
the rest of SGBNR users, so I'll try to explain it a little. I realize that 
the explanation below is going to be a bit long ---sorry, but probably I'll 
found no better way!

As you can see when you run SGBNR, parameters are grouped into three main 
sections: filter, edges protection, and masks.

* Filter

For the filter's order, you usually want small values, which can preserve 
smaller details. A value between 0.7 to 1.0 for order is appropriate for 
many images. A value of 0.4 generates the largest filter matrix and can 
yield very uniform backgrounds, but preserves very small image features, 
which sometimes isn't convenient. However, the filter's order is usually a 
non-critical adjustment.

The filter's radius is the main control available to govern the filter's 
strength. This parameter must be adjusted according to the filter's order. 
Generally speaking, low order filters require smaller radii.

The matrix size informs us about how much filtering strength we are going 
to apply. For instance, for 2700 dpi film originals, matrices between 15 
and 40 are usually required, depending on the importance, size and 
characteristics of film grain. For example, a filter of order 1.0 and 
radius 3.0 gives a matrix size of 41. The same size is achieved with order 
0.4 and radius 1.6. But the 0.4 filter is more conservative. The higher the 
order, the more averaging effect.

The iteration parameter is new to version 1.0.9. For mathematical reasons 
which are out of scope here, and as demonstrated by many test results, 
applying a smaller filter matrix several consecutive times is more 
efficient (in terms of better preserving low-contrast features) than a 
single pass with a stronger filter. For example, instead of a matrix size 
of 41, try two iterations with 19, or even three iterations with 13. Use 
this feature mainly for images with subtle low-contrast features.

* Edges Protection

Protection of edges plays a key role in SGBNR processing. For example, 
suppose we have an image of the Moon. You don't want the darkest features, 
as shadows on the bottom of craters, contaminated by surrounding bright 
areas. Similarly, you don't want extremely small bright features, as 
certain prominent peaks or thin crater borders, dimmed by pollution from 
neighbor dark areas. Edges protection helps protecting small detail and 
preserving contrast.

For both the dark and bright sides of edges, you have threshold and 
strength parameters. The most important parameter is by far threshold. Low 
threshold values are more aggressive when protecting edges from pollution. 
Each threshold must be adjusted by trial and error (don't worry: practice 
helps :-)). Start with the default values and preview. Try to reduce dark 
threshold. Too low a dark threshold yields undesirable dark artifacts, 
since it exaggerates the importance of extremely small dark features. The 
same is true for bright threshold: a too low value gives many small bright 
spots. It's unlikely that the default values (0.250) be too low, but 
sometimes that occurs. It's a good idea to preview some portions of the 
image with edges protection enabled/disabled; that helps evaluating the 
real effect of this mechanism. Once you've found a nearly correct threshold 
value, you can play with strength. Try previewing many different areas of 
the image. Areas with small dark features, with small bright features, with 
both of them, with none of them, and so on. E.P. is a beatiful creature.

* Masks

The masks protect certain areas of the image from excessive blurring during 
the low-pass filtering process. Masks in SGBNR work exactly like Photoshop 
masks: where a mask is black no change occurs; where a mask is white 
there's no protection at all. You can use two kinds of masks: uniform and 
image masks. Uniform masks are useful when we have to remove noise on the 
whole image, including bright areas. See, for example, a recent test 
processing by Jay Potts with an image of Hyakutake. We have placed our own 
version of the Jay's test on our website:

http://www.pleiades-astrophoto.com/software/SGBNR-test-potts/sgbnr-test-hyakutake.en.html

In this test you can see how uniform masks are the only way to get rid of 
noise on the comet's head. With version 1.0.9 we've got even better results 
with iterative filters:

http://www.pleiades-astrophoto.com/software/SGBNR-test-potts/108_51-4.tif
http://www.pleiades-astrophoto.com/software/SGBNR-test-potts/108_51-5.tif

The above images are, respectively, the result of two and three iterations:

108_51-4.tif
Two iterations with the following parameters set:
Uniform masks = 0.75
Radius = 0.85 (matrix size = 21)
Order = 0.40
Dark sides threshold = 0.270
Dark sides strength = 0.50
Bright sides threshold = 0.170
Bright sides strength = 0.15

108_51-5.tif
Three iterations with the same parameters as above, except:
Radius = 0.60 (matrix size = 15)

Image masks are much more restrictive than uniform masks: they protect 
bright areas and allow more noise removal on dark background regions. 
Starting with version 1.0.9, the usage of image masks is much easier 
because only three parameters are required to define the masks for the 
three R,G,B channels: shadows and highlight clipping values, and midtones 
balance. The clipping values work like Photoshop's auto levels. See the 
documentation for SGBNR version 1.0.7, under the "Auto" button. The 
midtones balance works just like a histogram's midtones balance. Again, the 
1.0.7 documentation will be of great help.

For the Matt's LMC image we used image masks. We found that uniform masks 
are not appropriate for this case, the main reason being that there is no 
noise to remove in the brightest areas of the image, especially the bright 
nebulas: uniform masks "wash" them too much.

And that's all folks! Sorry if I've bored you too much, but hope it helps!

Sincerely,
______________________________________
Juan Conejero, Pleiades Astrophoto Lab
skycad@ctv.es
http://www.pleiades-astrophoto.com/en.html

At 18:46 22/05/2002 -0700, you wrote:
>Juan,
>I too am very impressed with your LMC processing.  Thanks for the detailed
>explaination of your control of vignetting.  Now if I read it four or five
>times and try some of it, I may begin to understand it.  I do hope you can
>automate the anti-vignetting process.
>
>I downloaded your SGBNR settings used for LMC and notice they are widely
>different from the default settings.  At the risk of asking Rembrandt how to
>paint masterpiece art, how do you know what settings to adjust in SGBNR, and
>in which direction?  I'm at the trial and error stage.
>
>Don
>
>--  APML Archives at <http://astro.umsystem.edu/apml/>  ---
>              Unsubscribe at <majordomo@seds.org>



--  APML Archives at <http://astro.umsystem.edu/apml/>  ---
             Unsubscribe at <majordomo@seds.org>