deconvolution

[monoxide]

hi guys,
i've been trying to find a deconvolution for dummies type explanation but havent come up with the goods yet.

can someone please explain to me exactly what its doing and what im aiming for when using it?

also would it be better to split my stacked fits into R G B and go at each channel separately?

thanks.
Tony

[Alchemy]

i use the one in WIN4AIP as it works the best for me.

with this one it selects the Point spread function (how blurry the stars are) using a click on small stars ... mine is typically 2.4 to 2.6 this will vary if you use a high mega pixel camera it could be as high as 5

the deconvolution process is a bit like a sharpening, a little bit at a time. I use typically 40 itterations . as each itteration is done it is displayed. at some point it will all go ugly as its overdone, this is typically where i stop.

use a small selection for testing , preferably a reasonably contrasty bit , before you do the whole image as it takes a while.

the smoother the image the better the result, as this proccess will make noise stand out even more.

i do it reasonably early in the process often just selectively blended back into the pre- deconvolved image ie dust lanes in a galaxy .

[monoxide]

thanks Clive,

i think the PSF was set too high as it just kept cooking the image after a few iterations

[gregbradley]

The word deconvolution means to reverse the effects of convolution.

Convolution here means the process of averaging pixels based on its neighbouring pixels.

So deconvolution is removing the averaging or blurring aspect. Not sure where the averaging process is occurring in an image that is taken at 1x1 binning but thats what the word means.

As mentioned it is similar to sharpening but it seems to recreate the star image and shrink it more than simply sharpen.

I find doing it on all LRGB subs produces good effects. Especially on galaxy type images or close up type images where the stars can be large and distracting.

On widerfield images it isn't worth doing.

I click on several stars to find the one that is tightest and then run 40 to 60 iterations (repeats). It takes a while - like 5 minutes on a large image or longer.

Greg.

[netwolf]

Its been a while but I think this is how it went...

Convolution is simply a process to work out the output of a system. If X is the Input and H is the system response than by convolving X with H you get the output Y. If your Stellar Image is Y and the you know the H for your optical system then you can deconvolve and get X. In this case X would be the Image of the object without the affects of the Optical Train.

Usually H is the unkown but an estimate is used. A better way is to acutally work out H, often refered to as Point Spread Function. You can actually get a good Estiamte of H by capturing an Image of a Star or if there is a Star in the field using that as an estimate for H. For Planet Deconvolution it would be good to get a Star in the FOV while Imaging the planet and use the Star's Image as the H to deconvolve.

In a simple way Y = X * H if you know Y and H you can work out X.
Note "*" is a convolution not multiplication there is a difference.

H in Signals and System is referred to as the Response of a System to an Impulse Signal. The Filter response Curves for a Filter is a the Impulse response for that Filter. You could reverse the effect of the filter by deconvolving this response with the Output.