Deconvolution comparison

[telecasterguru (Frank)]

I have taken Marc's advice and run a CCD Stack deconvolution program over my image of Centaurus A. 1px at 35 iterations. I also sharpened the original LRGB files and used dust and speck filter in PS.

The image on the left is the deconvoluted image and the image on the right is the original image.

Doing the curves in PS on the deconvoluted image was more difficult. It seems that some of the data is lost in the deconvolution process.

Let me know what you think of the two images.

Frank

[TrevorW]

Original better

[multiweb (Marc)]

Looks good Frank. In my experience you won't lose details with a slight deconcolution but you will in PS with sharpening if you're not careful whethere you use highpass filter or another method. Also the dust and scratches noise filter has to be used with some kind of threshold. In your pic I'd say a minmum of 20 or you'll kill details quick as well as all those white specs. When you apply filters it's good to create a black mask on the same layer and only paint white the area you want to be corrected so the whole image doesn't get modified. Finally when saving a 16bits scaled TIFF from a raw file in CCDStack, make sure you lower the background slider right back or you'll black clip your picture histogram, which is the case in your shot.

[jase (Jason)]

Deconvolution can wield its magic but the output can vary considerably based on the quality of data. A strong data set will handle deconvolution far better than a weak data set. If your data has issues such as hot pixels or other anomalies, deconvolution will only magnify these issues. CCDStack's PC algorithm is one of the best around. Though there are plenty of other options to explore (http://www.princeton.edu/~rvdb/image...aximDLnew.html for example). I do however go by a general rule...if you're looking to maximise the quantity of nebulosity, don't use heavy deconvolution across the entire image. In most cases, you'll find deconvolution will suppress the faint wisps. I typically go with a PC of 30 iterations across the image to simply tighten up the stars. A second (and third if warranted) heavily deconvolved layer is then introduced to further enhance specific features using masks. The more control you have over the data, the better the result you'll be able to achieve.

[scopemankit (Chris)]

Deconvoluted image lacks the clarity of the original.