strange behavior in planetary images

[luzestelar]

Hi guys,
Yesterday I was processing planetary images on two different computers by using the deconvolution algorithm and same PI version 1.05.06. In one of the computers the Van Cittert (same parameters in both computers) creates gridded images, as you can see in the sending pictures

What is causing this problem?

Regards,
Jose Fernandez
www.luzestelar.es

[luzestelar]

Hi guys,
Yesterday I was processing planetary images on two different computers by using the deconvolution algorithm and same PI version 1.05.06. In one of the computers the Van Cittert (same parameters in both computers) creates gridded images, as you can see in the sending pictures

What is causing this problem?

Regards,
Jose Fernandez
www.luzestelar.es

[Cheyenne]

Color management settings and different embedded ICC profiles?

[luzestelar]

I've tried but has not solved the problem.
Thanks for your help

Jose Fernandez
www.luzestelar.es

[Juan Conejero]

Hi José

Could you tell us some specifications? I am particularly interested in:

- The software(s) you've used to align and stack the set of video frames
- The exact processor brands and models for both machines
- Amount of RAM installed
- If possible, brand and model of both motherboards
- Brand and model of both video cards
- Running operating systems

I may ask you to let me work with your original images. As far as I can tell, PixInsight is a deterministic system —for now, we don't use much quantic computing <g>. So either you have discovered a really odd bug, or something very strange is happening in your machines.

Thank you for your report.

[Carlos Milovic]

I bet you "quantice" something (For instance, spatial coordinates jajajajaja)

[luzestelar]

Hi Juan,
The images was stacking with Registax 5 in color tiff with 16 bits  deph. Both machines running XP with SP3, with DualCore and 4 MB Ram. One computer have Nvidia video card (in that i can see the problem) and another one have Ati video card.
I can see the problem only with the Regularized Van Cittert, the Richardson-Lucy work properly in both computers.
Attached i send you the Registax original image in jpeg format

Thanks for your support,
Jose Fernandez
www.luzestelar.es

[Cheyenne]

Totally misread the original post..

Does PI use any Windows DLLs as part of the image processing steps? (e.g. reading or writing the TIF formats, etc..).

Might want to check the original images that were loaded (e.g. swap the files and process them)

[Carlos Milovic]

File I/O is defined for each format in a separate module (.dll file in the PCL/bin folder). As long as I'm aware, this is cross-platform. Also, all the processes are defined in those modules, wich are cross-platform too. They do not use Windows' code or call, other than memory management.

[luzestelar]

Hi Juan,
I inform you that the same problem is observed in images generated by the K3CCDTools (tiff 16 bits), as you can see in the attached image, when you use the Regularized Van Cittert algoritm.

Regards,
Jose Fernandez
www.luzestelar.es

[Carlos Milovic]

Hi Jose

Could you upload the same image (last Jupiter) before the deconvolution?

Certainly this is very odd.


PS: A "workaround" to make the images workable again: disable the first layer with wavelets.

[luzestelar]

Hi Carlos,
Due to the pictures size I sent them to your hotmail.
Thanks for your help

[Carlos Milovic]

Thanks. I'll take a look at them tomorrow, and see if I can reproduce this behavior.

[Carlos Milovic]

Sorry. I had hardware problems with the computers running the latest PI core. Today they were solved, so tomorrow I'll take a look at them. Again, sorry for the delay.

[Carlos Milovic]

Hi Jose

Just examined your pictures... and I got the same results as you, a grid (working the the greenish jupiter). But, I don't think that this is a problem with the Van Cittern algorithm. The results are very similar with the regularizated and standard algorithms. Also, I got the same pattern with a very low noise modelation with the Wiener filter. If you examine the pixels at a great magnification, you may identify a subtle pattern on the original data. So, this leads me to think that this is only a side offect of the huge enhancement that has been done to the high frequencies.

I tried a previous noise reduction with wavelets, even deleting the first 2 layer (with a small scale function), but did not help much in terms of the results.

Don't know how to solve this... maybe if you oversample the raw data, and try to stack that... or, by the other hand, just avoid using the Van Cittern algorithm, and try to get a good result with RL, wich seems to work at a larger scale. Also you may try playing a bit with wavelets.

PS: BTW, after a mild noise reduction with GRAYC, the wiener filter does a pretty good job. Not as aggressive as Van Cittern, but a nice compromise.