Sharpening Strategy - Capture, Creative & Output Sharpening

[rhadorn]

I am presently trying to cope with the sharpening strategy and came across the interesting discussion started last January by Colin (When/How to Best Sharpen a Digital Photograph).

My present workflow consists in developing raw files with Capture One and post processing them in Lightzone. My general strategy is to develop 'flat' (checking the 'linear' option instead of keeping the 'Standard Film' option) and to apply a tonal curve later in Lightzone, together with local contrast enhancements using an USM tool at various radius - say, 8, 24, 64 as an example. Of course, there is a capture sharpening in C1 of the 300 /.7 / 4 type. I always apply small radius before large radius, to avoid strengthening halos.

My understanding of the creative sharpening mentioned in the discussion is that it has more to do with local contrast enhancement than with actual sharpening, starting with a radius of 2.3 for example at an amount of 53/500. Whether it is sharpening or contrast enhancement depends actually on the subject. If the latter consists on large homogeneous surfaces, it will be a sharpening. If you have a lot of grass in the foreground, it will be local contrast enhancement...

The issue I am struggling with concerns the necessity of being selective, when sharpening or enhancing contrast. I usually exclude the sky, at least at small radius USM, but sometimes i am too lazy to do this and there it happens, I get strong dark and light borders at the frontier between that rock and a brilliant haze in the sky. Selective sharpening was also mentioned in the discussion, this is actually a critical point (time factor multiplied by the limits of the computing resources at your disposal...)


But I certainly will be lazy again in the future and wonder why the USM algorithms did not evolve to cope with that difficulty. I came across various masking techniques, applied either to sharpen selectively borders or on the contrary to exclude identifiable borders from sharpening. Would it not be easier to add a parameter, namely an upper threshold, avoiding sharpening where the differences in luminance excess a certain value, say 100 or 120? It would be much easier to avoid sharpening artefacts where they appear first, namely at luminance frontiers, which are clearly identified anyway? Lightzone does not have the flexibility of PS, but you know, limitations sometimes also makes one to be creative.


Yesterday, I was confronted to an output sharpening issue. I print using QImage, a specialized program, so I let it output sharpen my pictures to its taste; the default setting is actually 'moderate'. Now, I had a picture, which I re-edited several times to moderate those sharpening artefacts at the transition between rock and sky. They did not go totally. At last I made an additional print setting the output sharpening by QImage to zero and there it happened: the double halo vanished. So, even a moderate output sharpening can have devastating effects if it is not closely controlled. Actually, QImage tells nothing about what values are applied when you select that 'moderate' sharpening...

Most of my printing from a 10 Mp camera is on A4 at 360 or on A5 at 720 PPI, meaning native resolution. If I understand correctly Colin's comment on January 21st, I don't actually have to output sharpen. Output sharpening is important after downscaling.

Is it also after upscaling, say to 250 PPI?


Reto

P.S. I already ordered Fraser's gold mine, thanks for the reference!

[Colin Southern]

My understanding of the creative sharpening mentioned in the discussion is that it has more to do with local contrast enhancement than with actual sharpening

The two inter-act; sharpening increases contrast around edges whilst local contrast is also about tones with relation to their surroundings. Obviously the requirements are going to vary depending on the image.

Personally, I regard local contrast as being on a larger scale than sharpening (rightly or wrongly); capture sharpening corrects the softening caused by the digitisation process - the anti-aliasing filter - and the demosaicing process giving the image a pleasing clarity when working at 100% views (but really doesn't do anything as far as the average print goes as the difference is too small to be seen) - content/creative sharpening is done a lower levels of magnification to give the image a sense of "snap" or even a slight "3D" illusion, but for local contrast enhancement I usually use tools like dodge/burn to enhance the contrast over a defined local area (similar to the effect one might get selecting an area and applying a levels layer and then playing with the clipping controls, but without having to worry about masking) (a good example would be running a burn tool - set to shadows - over the branches of a tree to enhance their local contrast with respect to their background).

The issue I am struggling with concerns the necessity of being selective, when sharpening or enhancing contrast. I usually exclude the sky, at least at small radius USM, but sometimes i am too lazy to do this and there it happens, I get strong dark and light borders at the frontier between that rock and a brilliant haze in the sky. Selective sharpening was also mentioned in the discussion, this is actually a critical point (time factor multiplied by the limits of the computing resources at your disposal...)

In Photoshop a quick and dirty fix for local issues like that is to use the history brush to selectively roll back some or all treatment of certain areas - not sure if your package will have any equivalent tools.

But I certainly will be lazy again in the future and wonder why the USM algorithms did not evolve to cope with that difficulty.

Who knows

Most of my printing from a 10 Mp camera is on A4 at 360 or on A5 at 720 PPI, meaning native resolution. If I understand correctly Colin's comment on January 21st, I don't actually have to output sharpen. Output sharpening is important after downscaling.

No. Output sharpening is still needed in those circumstances. I'm quite lucky in that I pretty much only work with quite large images (a small image for me is one where the longest dimension is 22 inches) so for me the effects of content/creative sharpening pretty much just carry through to printing without anything further being required. Opinions differ, but personally, I doubt you'll see much (if any) difference at resolutions higher than 180PPI at normal viewing distances - that's what I print at (and I'll even go down to 100 PPI if desperate).

Hope this helps

PS: I'm not sure how equivalent USM settings are between our respective programs, but (almost without exception) I use 300/0.3/0 for capture sharpening on ISO 100 images - I would have thought that settings like 300 /.7 would have played havoc with smooth textures like clear sky and skin tones (although a threshold of 4 would have decreased the effect a lot). My suggestion is to leave threshold at 0 unless you have a very noisy/high ISO image.

[KentDub]

But I certainly will be lazy again in the future and wonder why the USM algorithms did not evolve to cope with that difficulty. I came across various masking techniques, applied either to sharpen selectively borders or on the contrary to exclude identifiable borders from sharpening. Would it not be easier to add a parameter, namely an upper threshold, avoiding sharpening where the differences in luminance excess a certain value, say 100 or 120? It would be much easier to avoid sharpening artefacts where they appear first, namely at luminance frontiers, which are clearly identified anyway? Lightzone does not have the flexibility of PS, but you know, limitations sometimes also makes one to be creative.

I can't help you might with Lightzone -- but it may have something equivilant. Photoshop has 'Advanced Blending' options which can accomplish, rather elegantly, what you are asking for. This, of course, requires that the USM was run on its own layer. Take a look at what kind of control you can get for layer blending and let us know

[rhadorn]

History brush - This is a selection (or de-selection) tool I suppose. This is precisely what I start doing more systematically: sharpen selectively. It means a lot more work but also a bit more satisfaction with the result. Lightzone offers flexible polygons and curves with an easy feathering for selection.

Advanced blending modes in Photoshop - I compared the blending modes in Paint Shop Pro (a kind of copy of PS) with those I have in Lightzone: they are very similar and the selection of another mode than 'normal' does not help. On the other hand, PSP lets one define blend ranges, and so does Lightzone (color and luminance selection). I am using this one more and more; associated with the selection of parts of the image, it makes allows a very differenciated sharpening of the image. What I try to avoid is the extreme darkening of some black spots expressing small shadows.

What I dreamt of is something more 'automated', supposing an analysis of the image, allowing a self-controlled reduction of the sharpening effect where adjacent pixels are very different instead an amplification of the difference, which is the way USM works usually. Such an algorithm would also avoid the extreme darkening or lightening of small zones, which just happen to be near very light or dark zones (the size of the zone depending on the radius at which I sharpen or push local contrast). As long as we don't get this, I will do the selection myself.

Local contrast - I also use the local correction of the tone curve. A second means I use is the USM at high radius (50 - 500) but low amounts (15 - 7). Using several radii at a moderat amount has quite an interesting impact and should not build artefacts. Depending on the image, you will apply this to the whole image or only a part of it.

Capture sharpening - You may be right, Colin, that 300 / .7 is too harsh. It is difficult to control this on screen. There are a lot of recommendations around. Capture One proposes default values based on your camera and ISO value. A competent user of the program I discussed with recommends to use lower values than the ones proposed by the program; where the program proposes 180 / .8 / 1.0 he corrects to 100 / .6 / .9. Another source sets the fork as follows: 60-90 / .4-.8 / .6-1.5. A lot of experience is needed to set up a good practice, given the differences between cameras and data quality depending on ISO values.

Can you pleas explain once more the utility of output sharpening while upsizing an image? I don't clearly understand your comment on this; you write that output sharpening is always needed but underscore that creative sharpening is preserve anyway when upsizing - well, this is my understaning of your comment.


Thanks to both for your support

Reto

[KentDub]

Can you pleas explain once more the utility of output sharpening while upsizing an image? I don't clearly understand your comment on this; you write that output sharpening is always needed but underscore that creative sharpening is preserve anyway when upsizing - well, this is my understaning of your comment.

When you upsize an image the creative sharpening is preserved, softof. It is preserved proportionatly to the reset of the image. When you apply output sharpening, the areas you creativly sharpened are still that much proportionatly sharper.

Sharpness is lost when the image changes size - bigger or smaller. Imagine two peices of grid paper that do not line up, with the 'boxes' different sizes. Imagine random boxes filled in on one, then try to figure out how to map that to the other peice of paper. A perfectly black box next to a perfectly white box on one sheet will not be perfect on the other, as detail it either lost (factoring in surrounding pixels, it will lose contrast), or interpolated (black->grey->white).

Output sharpening is device-dependent. This means save/backup the working file and/or apply this as a non-destructive edit. Two primary uses for output sharpening are sharpening for print, or the web. We sharpen for print because inkjet nozzels naturally blur things when they spray ink on the paper. To counteract this, we 'oversharpen' (output sharpen) for the printer. Lots of times we resize the image for print, which alone is reason enough to resharpen the image - If the image is close enough adjusting the DPI of the image will yield better results, resizing should be avoided if at all possible. For the web we downsize the images dramatically for two reasons: We don't want people to have to zoom out to see the image on their screen, and two, we need the filesizes low so people don't have to wait a long time to see it. Because we resize it, we need to resharpen (output sharpen) the image as again, detail and contrast are lost when the image is resized.

Hope this helps

[rhadorn]

Hope this helps

Yes, it does, indeed. Thanks.

Reto