Source vs downsampling vs output sharpening?

[Marsu42]

I have always been puzzled by this and never found a conclusive answer in the general photography forums: When do I apply what amount of sharpening?

The possibilities are:
1. before downsizing (esp. Lightroom/PS with ACR since it has access to the original raw bayer pattern and can adapt to the camera type)
2. while downsizing (like EWA Lanczos3Sharpest: http://www.imagemagick.org/Usage/filter ... downsample)
3. after downsizing (with -adaptive-sharpen).

* Is there any good reading available on this?
* Any recommendations, for example if I can simply skip step 2. in favor or 3. or vice versa?

Thanks!

[snibgo]

Some people say you should do a little sharpening as the first operation, before shifting tones/hues etc. Until recently, I did this. Then no more sharpening until the image is at the final size or, if you prefer, as part of the downsampling. Personally, I downsample then unsharpen as a final step, as I find that more controllable. Different images at the same size need different amount of sharpening.

I don't think there are conclusive answers on this topic.

Some good websites:
http://www.computer-darkroom.com/home.htm
http://www.cambridgeincolour.com/
http://www.luminous-landscape.com/index.shtml

[Marsu42]

Some people say you should do a little sharpening as the first operation, before shifting tones/hues etc. Until recently, I did this.

Meaning you don't do it anymore ... why's that? Too much processing power for too little of a result? Or artifacts in certain constellations vs. pure output sharpening?

Different images at the same size need different amount of sharpening.

So you look at each picture's pixels individually and then decide what sharpening amount to apply after downsizing? But if you're running a batch operation - what would you rather do: sharpening with the downsampling algoritm or adaptive-sharpen afterwards?

[snibgo]

Meaning you don't do it anymore ... why's that?

Ah .. I was afraid you might ask. I have recently rejigged my workflow, automating a gamma and sigmoidal correction, and I needed the rawest values to judge this. Sharpening before this just confused me. Now I'm quite happy with those corrections, I might put the sharpening back in. But I'll need to test, to find where it should go. I might get time later this month. Or next month.

I'm not bothered about the processing power. It's a cheap operation, and I can make a cup of coffee while the computer is brewing.

So you look at each picture's pixels individually and then decide what sharpening amount to apply after downsizing?

Kind of. I grew up with B&W photography, where I would look at the negative and say, yeah, that will need 20 seconds, with 5 seconds burn in that corner. Digital is similar, but I have jpegs from the camera as proof prints, and the first batch run gives me rough 10x8s (so to speak), and I punch numbers into a file so this bunch of photos will get a relative sharpening of 1.0 while that other bunch will get 1.5.

But then the real fun starts: what images work, what don't, why? And curves. I've never made a photo that couldn't benefit from a tweaking of curves to emphasise this or tone down that. Often I'll apply the same curve to multiple images. (I do curves in Gimp, then read the Gimp data from an external program.) Then the more complex stuff: burning, dodging, hue shift, differential blur/sharpen, and off we go.