Can PI fix my STL-11K vertical bars?

[bob_franke]

Hello Everyone,

I have been working with the PI demo for about a month and things are starting to come together.  I have an STL-11000 camera and because I aggressively stretch my images, I am always fighting the problem with the CCD's vertical bars.  I have a PhotoShop fix, but is does not work very well.  In process of masking out the bars, it adds vertical artifacts to the brighter stars.

So my challenge to the PI pro's is can someone fix the bars with PI.  The below JPEG shows the bars in an Ha shot of M74.  This is a stack of nine dithered 30-minute exposures.
http://bf-astro.com/m74haScaled.jpg

Here is a zip file with nearly the full frame.  There are two images, a stretched TIFF and the original 32-bit FITS.  So it's a bit large... 48.6MB.
http://bf-astro.com/m74Ha.zip

Regards,
Bob
http://bf-astro.com

[Cleon_Wells]

Bob, I used Georg's CanonBandingReduction Utility to modify your image, this may not be the best way in PI to correct your problem but it does reduce the banding.
I first rotate the image 90 degrees and run this script and then rotate the image back and then ran this script again.
This Utility is located here, Script/Utility/CanonBandingReduction.
 Cleon

[georg.viehoever]

Yes, the script is one option. But: did you calibrate your images (bias, dark, flat)? If not, this should be done first.

Georg

[Cleon_Wells]

Bob, after Georg's suggestion, I forgot to mention that I should have checked the Protect from Highlights Box, use a SigmaFactor of around 1.30 for this image.
Cleon

[Nocturnal]

Good grief, that's data from a top of the line SBIG? Is that normal?

[mmirot]

Calbrated images from this camera should not have this artifact.

Max

[bob_franke]

@ Cleon:  Thanks for pointing out this utility.  It looks promising.  I just found a Kodak application note that says the vertical bars are cause by improperly aligned vertical clock pulses.  So now SBIG's excuses are in question.  They have always said the problem is with Kodak.

@ Georg:  I am a very experienced imager... of course the sub exposures in the image stack are calibrated.
BTW... Thanks for developing BandingReduction utility.  It looks promising and I may be back with some questions about how to use it.

@ Sander: The vertical bars are not normal, most STL-11000 cameras are OK.  If you are referring to the noise, that is normal for a single 30-minute narrowband image.  I usually stack at least 20 30-minute exposures.

@Max:  Of course this camera, or any, should not have this artifact.  However, many of the SBIG STL-11000's produce images with these vertical bars.

Regards,
Bob

[NKV]

Bob, these vertical bars always in same position or moving (lift-right) from picture to picture ?
I ask you because I see similar bars (but horizontally in my FLI PL16803) with bad power supply of my laptop. And these bars disappear if my laptop running at battery.
So, maybe your problem in electricity too?

[bob_franke]

Hello NKV,

I wish it were a simple as a PC power supply problem.  Here's the Kodak application note that explains exactly the vertical bars (16 pixels wide) that I have.
http://www.kodak.com/ek/US/en/VerticalTimingOptimizatioInterline.htm

With this document in hand, maybe I can convince SBIG to fix the camera.  Their service is usually very good.

However, this is getting a bit off the track of my original post.

Regards,
Bob

[pfile]

ugh, those kind of timing problems suck. hopefully these are very slow, derived clocks and they can use a state machine to insert those delays, otherwise it's pretty difficult to do right across temperature and voltage.

[Carlos Milovic]

Hi Bob

Since the vertical bars follow closely a sinusoidal pattern, you may take advantage of the Fourier Transform to eliminate that effect.

I made a simple experiment with the result attached.
The procedure is as follows:
- Extract the fourier transform (see FourierTransform process) of the stretched image.
- Create a new image with the same dimensions (use the active window). New Image process, under Image category.
- Apply the following Pixel Math expression to it: iif( (x() == 1920 || y() == 1350) , 1, 0)
This will create a black image with a white cross in it, in the center.
- Now, using the Cloning Tool, erase the center of the cross. Just a small section, completelly  removing the center.
This will be our mask. Note that the cloning step may need fine tunning to avoid artifacts. This means, you may get haloes in the image, specially from the galaxy. Too wide, you will not erode the noise pattern, too narrow, artifacts will appear.
- Select that image as a mask for the Magnitude component of the fourier transform.
- Apply one of the following processes (experiment with the results):
+ Gaussian blur
+ Median filter
+ PixelMath with the following expression: $T*$T
+ other process that may average intensities or attenuate them.
- Take the inverse Fourier transform, using the modified magnitude, and original phase.

This procedure is known as a Notch filter. I may write a more controllable and easy to use process to perform that, or even try an adaptative algorithm I read about some time ago. Let me know if you are interested, although I can't promise to be too quick to do that.

BTW, there may be some boundary problems with the filter following the procedure described above. This is a result of the zero padding performed by the FourierTransform module. Ideally you should expand the image and mirror the contents in the new space. You may also expand it with a constant value (the Image's mean value), and it would also work better than the first approach.
Also, you may correct any large scale artifact (gradient) with DBE. Even the artifacts generated by the first procedure. A little time consuming, bu this should get the work done.

[georg.viehoever]

Wow!  8)

[bob_franke]

That's impressive work Carlos.  There's no way that I could come to a solution like this on my own.  Knowing that the pattern is 16 pixels wide and spaced at 16 pixels, you can probably get an even better result by working with the sub exposures before stacking.

It looks like SBIG is going to repair my four-year old camera.  They have even offered to pay the postage and insurance costs.

If SBIG repairs the camera, they want to see an example image first; I won't need your proposed filter.  However, it would be a great addition to PI... as time permits.

Thanks and regards,
Bob

[Josh Lake]

Back with this old chestnut... does anyone else have the vertical banding issue on their STL-11000? I usually can't detect it, but I shot 5 hours of Horsehead data last night in H-alpha and can now clearly see it.

I looked over this thread but did not try Carlos' brilliant Fourier solution. I did some searching and found the SBIG's own CCDOps has an option for this.

Then I found something called CCDBand-Aid that claims to do the same but in 32 bit floating point mode: http://bf-astro.com/ccdBandAid/ccdBand-Aid.pdf

Problem is, even though I'm feeding it 32 bit FP images, it comes up with an error that it *only* takes 32 bit images. My images are definitely 32 bit!

Does anyone have experience with CCD Band-Aid or a 'best solution' to this issue? I'd love to fix this data.

[georg.viehoever]

Did you try CanonBandingReduction in the scripts section of PI? It has some success fixing horizontal bars as seen in Canon cameras. By rotating your images it may be applicable to vertical bars as well.
Georg