Deconvolution question

[cmarcus]

Just getting started with Deconvolution and I need to know which channels to do it on.
If I have LRGB data, do I just do decon on the L? Or on the RGB combined as well? Or on each separate channel, R,G and B.
Likewise if doing narrowband, HA, SII, OIII + HA luminance do I do it to each or just the HA luminance?
TIA for any input.

[Stu]

I've never been able to get deconvolution to work that well as I tend to be undersampled so I usually skip that step, but the times I've done it I do it only on the lum channels.  Not the color. 

[cmarcus]

thanks for that, I've only done it a couple of times, but have had some success with it. Seemed better when I did it on an HA to be used as luminance, than when I did it on a genuine luminance of an LRGB composition.

[Warhen]

 Deconvolution should be reserved for luminance when available. If working with one-shot color data, be sure that the lightness channel is selected as the target. As a rule of thumb, never deconvolve chrominance.

[Physicist13]

why only lum?? isnt deconv a means to make non-circular stars circular? should apply to all channels??
Patrick

[Dimitris Platis]

Deconvolution is not a way of making circular stars. It is used for bringing up details, sharpening, enhancing. A side effect is making stars sharper.
The luminance chance is responsible for these effects, contrast.....so u dont need to apply on RGB. In fact, many times its better if RGB is more blurry so u can avoid pronounced  noise

[Physicist13]

disagree. Its applying a SPF and doing a mathematical deconvolution.
Patrick

[dnault42]

I think the reason that most people do deconvolution on the luminance only is that the RGB data is allowed to be noisier.  Usually this is intentional and is done either through binning which is scaled up increasing per pixel noise relative to the luminance or through less exposure time.  In general this is find as the luminance is used to create the detail and more aggressive noise reduction can be used on the chrominance information.  The problem is noise is extremely disruptive to the deconvolution algorithm making it very difficult to get any improvement from typical RGB data collected as part of an LRGB image.

Running deconvolution on the luminance only can lead to problems like color bleed/halos around stars or high contrast objects but how noticeable this is depends on how aggressive you are with the deconvolution and how protected the stars are in the first place (since we typically run with a star_mask for ringing control).

Regards,
David

[Warhen]

I think you'll find Patrick that in practical application, it's the luminance that drives the appearance of the stars, so just as luminance will dictate the contrast and sharpness of an image, so can chrominance generally be spared deconvolution. Patrick/Dimitris, you're both correct. A PSF is used to de-blur the stars, and yes as they're corrected, they tend to be rounder, smaller, and sharper. Nonstellar detail is also enhanced by this de-blurring effect.

[danlr46]

The reason depends on human visual system. Contrast sensitivity is better on luminance than color.
This is the reason why video encoding follows the rule 4:2:2 that means 4 samples for luma, 2 samples for chrominance channels.


Therefore improved luminance channel is sufficient to have better rendering.


Daniele