Jupiter image files to test your astra image skills on

[davidpretorius]

Hi Guys

http://www.precons.com/iceinspace/gallery/jupiter_060426_1147/

grab some of the .tiff files and then split in astraimage and then have a play with deconvolution and unsharp mask and fft etc.

there are various stack sizes and wavelet settings.

Ask questions and post your results to see how different deconvolution works?

[janoskiss]

there are quite a lot of files there. pick one or two for us Davo.

[Robert_T]

Dave, is FFT (what is it?) something that's available on the mid-range Astra Image 2.0 or only on the full blown version?

cheers,

[janoskiss]

FFT = Fast Fourier Transform. It provides an efficient way or performing a convolution operations (which in turn are used in deconvolution algorithms).

[davidpretorius]

in the demo version, to the right of VC deconvolutions

[asimov]

Cheers Davo, just what I need...more decon practice! :D

[asimov]

VC deconvolution 5 iterations/0.6 curves FFT edit on R & G. Not the result I was looking for...Back in for another go! :D

[janoskiss]

I did not use Astra Image but here is my effort anyway. I am at home sick today so this was a nice way to pass the time, keep my mind off my pounding headache.

For the most part I used a piece of free software called ImageJ (http://rsb.info.nih.gov/ij/), and the FFTJ and DeconvolutionJ plugins. I use it in Linux but it runs on just about anything (it's java based).

For best results deconvolution should be applied before any other image processing operations. So I chose a this starting image (see first attached pic): 1__ST149_WV10_10_58_91_118_186.tif, because it roughly looks like it has the degree of blurring expected for the telescope used (which will blur out of details smaller than about 4-7 pixels depending on colour), and does not show any visible signs of processing, i.e., sharpening (unlike e.g. 1__ST149_WV10_10_58_138_293_455.tif , which is a bit sharper but shows a bit of a white halo around the moon shadow).

For the deconvolution operations I estimated the point spread function (PSF) based on the following parameters:

* diameter of Jupiter on image = 412 pixels (measured with kruler off the screen, +/- 4 pixels)
* angular size of Jupiter = 40.1 arc sec (assume picture taken on 25/06/06)
=> we have pixel size = 0.09733 arc sec
* Telescope aperture = 250mm (10" Dob)
* Central obstruction = 65mm (info about 10" Dob in previous posts on IIS)
* wavelength = 475nm (blue), 510nm (green), 650nm (red)

I wrote a short C++ program to calculate the PSF with Bessel function J1 from the GNU Scientific Library (http://www.gnu.org/software/gsl/). I generated 3 PSFs, one for each colour. (See attached image - PSFs are shown at four times actual size and gamma adjusted to show ring structure.)

Decomposed the image into RGB channels and deconvoluted each using the corresponding PSF. The FFTJ plugin is a one-step direct method of deconvolution. There is a "regularisation parameter" that needs to be chosen appropiately (for best compromise between noise and sharpness). I chose these values by inspection: red: 0.002, green: 0.00005, blue: 0.001, which gave good detail plus a bit of speckle noise.

Then:
- RGB merge
- Despeckle: ImageJ has a very good parameter-free despeckle filter and cleans up the deconvolution noise very nicely.
- Level tweaks: black for each channel separately set to darkest pixel on moon shadow, white to global max (all channels together)
- Gamma: 0.67 0.70 0.60 (RGB) - by eye
- correct for slight purple-green fringing by realigning colours: shift red channel 2/3 pixel up and 1/3 pixel right (blow up image + move + shrink image) - see attached pic.

Result shown in last attached image. I'm quite pleased with the result for my first attempt at this sort of thing. :)

PS. I kept getting this strange error when trying to open the original tiffs with any image editors I tried: unknown field with tag 317 (0x13d) encountered. :shrug: Fixed with a pass-through raw2tiff.

[Harpspitfire]

bit off topic here, but is there any download for a deconvultion file anywhere without having to have it in a whole processing program? i tried iris and couldnt even figure out the program, something else (i forget) but it had the L-R decon, van client?, and another kind- ended up with jup images that werent any better at all- thx, john

[C.]

What a good idea for those of us who haven't started imaging yet. I just used my regular image program and had a play. Thank you for the opportunity David. I hope I can load it ok.

Janoskiss I hope you are feeling better now.

[janoskiss]

John, not sure what you mean. If you are after dedicated deconv software, they tend to be highly specialised and more complicated than integrated deconv routines in imaging programs.

I uploaded the PSFs I used for deconvolution to http://www.iceinspace.com.au/uploads/PSF.zip in case anyone else wants to have a go using these.

Once you have the PSFs, deconvolution is very simple to do with ImageJ (http://rsb.info.nih.gov/ij/). You need to get the program from the Downloads page, and the "FFTJ and DeconvolutionJ" plugin from the Plugins page. To deconvolute a greyscale image:

* open greyscale image to be deconvoluted (use original 16-bit tiff)
* open corresponding PSF file (also 16-bit tiff)
* increase canvas sizes to nearest power of two (if they are not already, in this case 512x512; use Image -> Adjust -> Canvas size)
* Plugins -> FFTJ -> DeconvolutionJ:
. - select blurred image to be deconv'ed and PSF image
. - enter a value for regularisation parameter or leave at default (can experiment here for best results)
. - you might like to do a Process -> Noise -> Despeckle on the deconv'ed image

One thing to note about the PSFs here is that they assume that the diffraction limit of the telescope is the main source of blur in the image, which is likely incorrect. Other sources of blurring, e.g. due to poor seeing, can be thought of as convolution operations too, and can be corrected for using deconvolution. Then one would need to estimate the PSF from the blurred image itself.

[asimov]

You certainly know your stuff Steve. I tend not to look into the technical side of it too much; I know next to nothing of the terms & so forth, & basically have no concept of what I'm doing. I do like to muck about in Astraimage to see if a better final image can be derived from it. Great job having never done it before ! :) Your effort is a lot smoother looking than mine.

[davidpretorius]

thats awesome steve, excellent stuff!

more stuff to read up on!

[Harpspitfire]

thx for the info steve- after looking at everything thats way beyond my processing skills!- i was just hoping to get a better final image- guess ill just have to play with photoshop for now- john

[asimov]

Well if your anything like me John (& it sounds like you are) get yourself Astraimage & just have a play around. http://precons.com/iceinspace/gallery/ai25max.zip

[asimov]

BTW nice job C :)