How much sharpening to apply?

[Lee_P]

I like my pictures to be very sharp -- enough to cut your finger on -- although atmospheric limitations mean I tend to overcook sharpening during processing to compensate. In particular, I'm pretty heavy-handed with PixInsight's UnsharpMask. What I'm wondering is, are there any tips or techniques to judge how much to apply? Or how to know if you've gone too far, other than eyeballing it?

For example, I used to apply too much noise reduction. I then picked up a tip from @vlaiv about keeping an eye on the smallest pin-prick stars in an image. If I wipe those out using noise reduction, then I've gone too far. Are there any similar tips for sharpening?

[Fegato]

I can't really give any tips, as I've never really happy with sharpening on DSOs. I've used UnsharpMask on the moon and planets, but rarely on a DSO image. I guess I'm mostly shooting fairly wide field, and prefer the "nebulous" look to risking production of artifacts and yo-yo-ing between sharpening and noise reduction. I do sometimes see an image that looks sharp and good with it, and think wow - wonder how they did that!

[vlaiv]

Here are some tips that might be helpful.

1. watch out for noise. In fact - SNR is different in different parts of the image and you can't sharpen the whole image the same if you want to get the most out of your sharpening. Best approach that I've found is as follows (Gimp but can be adopted to other software):

- create copy of the image

- sharpen that copy until you are satisfied how bright parts look. This will inevitably create a lot of noise in the background

- add layer mask to this sharpened image. Copy original image as layer mask. Invert mask as necessary - bright parts should show sharpened layer and dark parts should show original. Stretch the mask more then you did the image. With opacity slider for whole layer control how much of it is blended with original image.

- when happy - flatten the image.

Above approach can work for denoising as well. Only difference is that you want inverted mask - you want to show denoised version in dark areas where SNR is poor.

Signs that you've over sharpened:

1. There is dark ring around bright stars

2. You've reduced stars to bright single pixels

3. All stars in the image are at peak brightness (there is no natural variation of brightness among stars)

4. Noise obviously

5. You started to create posterization artifacts

 

[Lee_P]

Great, thanks!

[ollypenrice]

First thing, for me, is not to sharpen stars, so they always need excluding. Indeed, using star removal and replacement, I usually blur them by about 0.5 Gaussian.

Secondly, as Vlaiv says, don't try to sharpen weak signal and, because it's pointless, don't sharpen any regions with no small scale detail in them.

Thirdly, consider the scale of any sharpening. If you set USM with a higher threshold it will only operate on larger scales, working almost like local contrast enhancement. This can be good for structural boundaries in extended nebulosity, for instance. Small scale sharpening, however, works well on small features like galaxy detail. Basically, one USM set of values does not fit all parts of the image.

I do this in Photoshop because it means I don't have to faff about trying to get exactly the mask I want. I can just make a copy layer, sharpen the bottom and then erase the top with a soft brush where, and only where, and exactly where, I want it. The Select-Colour Range tool in Ps is also a very easy alternative to masking. (It is a form of masking but generated differently.)

Finally, I try not to forget what nebulosity means.

lly

[tooth_dr]

@ollypenrice good tips as usual Olly 👌🏻