Output Sharpening

[Boatman]

I have read many articles about input and composition sharpening. I think I have a pretty good handle on these subjects. However, output sharpening still seems like something of a mystery to me. My prints are good but probably have room for improvement, which I am always in favor of.

I did some research and came up with several pointers for output sharpening that I have developed into a workflow. The workflow makes sense, but honestly, I don't see any difference in sharpness from my previous printing technique. As a test I made three prints:

1. Cropped and printed at what looked like good sharpening for a normal image - no adjustments to the image size resolution (295 @ 11 x 14)

2. Cropped and sharpened to 70/1/10 using the following methodology

3. Cropped and sharpened to 200/.7/30 using the following methodology

Viewed from a foot away, you cannot tell these prints apart. Pixel peeping #3, you can see a little over sharpening. So, I'd say my technique is not working.


What's right with this? What needs fixing? What steps would yield sharper images?

This is the workflow:

Output Sharpening

Save document before further changes

Crop image to final print proportions

Re-size to print dimensions and 360PPI (Epson printer, Canon or HP may need 300) with re-sampling.

Copy all previous layers to a new layer, cont-alt-shft-E

Change layer type to Luminosity

Set screen view to 50%

Open up the Unsharp Mask
-have the preview panel set for 100%
-set radius appropriately for the frequency of the image (1 for normal images, .7 or less for very busy images)
-set the threshold to 10
-adjust the amount until the pixels begin to 'crinkle' slightly, then back off a little
-re-adjust the threshold to assure that even-toned areas of the image are not being over sharpened

Proceed to print using the printer settings appropriate for your printer and paper

When complete and the print is printed satisfactorily, save the image in a folder for final print files using 'save-as' and a name that will allow you find the document easily in the future.

Note on the print: file name, paper type, printer model, ink type, date

[Geoff F]

Just a couple of thoughts, Homer.

I think you mentioned that you are sharpening as Lumonisity blend layers. This was popular a little while ago but, as far as I can see, the general consensus today seems too be that 'Normal' works just as well for general use and Luminosity blend mode is better for problem images as the effect is less pronounced.

Also, you use a lot more Threshold than I do. My settings under similar circumstances are more like 100/1/2 max and a little less for noisy images, say 80/0.8/2.

And for internet images, I usually apply the same amount of final sharpening before downsizing then add a little 'downsize correction sharpening' of around 50/0.5/0.

[The Blue Boy]

Hi Homer,

We see this sort of question quite often here, so I'll refer you to this older thread, When/How to Best Sharpen.

I'd also recommend reading a few books by the late Bruce Fraser. The current ACR sharpening algorithm was designed to follow his workflow. I've found it invaluable!

[Colin Southern]

Hi Homer,

We see this sort of question quite often here, so I'll refer you to this older thread, When/How to Best Sharpen.

I'd also recommend reading a few books by the late Bruce Fraser. The current ACR sharpening algorithm was designed to follow his workflow. I've found it invaluable!

I also posted a bit more about sharpening here.

The industry standard reference text is "Read World Image Sharpening" by Bruch Fraser and Jeff Schewe (2nd edition).

Homer - I also typed up a longer reply from home this morning, but got an urgent call, and had to run without finishing it -- I'll finish the post tonight for you.

[Boatman]

I've re-read Jeff Schewe's piece, "Image Sharpening - Make It Really Sharp". In that article he covers a number of sharpening issues and describes work flows for generalized edge-mask sharpening and output sharpening. I find the edge-mask sharpening works well. The output sharpening does not seem to help.

Jeff recommends resampling the image to match the printer. I've seen this recommended in several places, but others state this is unnecessary. With my Epson 1400, apparently the ideal setting is 360ppi, but I don't see any difference with this at all.

If I do resample, what is the impact of the drop-down options: Nearest Neighbor, Bilinear, BiCubic, BiCubic Smoother and BiCubic Linear?

[Geoff F]

Personally, I have found that straight forward Bicubic is as good as any.

Nearest Neighbour or Bilinear can be a bit on the coarse side.

It was recommended to use Bicubic Smoother for upsizing and Bicubic Sharpener for downsizing.

I don't do substantial increases in size, since most modern cameras have an excess of pixels for most uses but using Smoother may be something to consider in extreme conditions.

Bicubic Sharpener, in my opinion, gives an auto sharpening which may not suit the image in question and I didn't like the results which I was achieving. So when doing substantial decreases, like resizing for the internet, I just use the standard Bicubic then apply a little Unsharp Mask to suit each photo.

[Colin Southern]

Hi Homer,

Sorry, I didn't get around to posting that info I started a few days ago (me bad)

Quick recap on what sharpening is all about ...

... in a nutshell, increasing edge contrast.

Capture sharpening is done on a full-resolution image, and is necessary to overcome the effects of the anti-aliasing filter - digitisation process, and demosaicing processes, so not surprisingly, it only needs a small radius (0.3 pixel is pretty standard), but a fairly agressive amount (I use 300%). We don't want to sharpen noise, so thresholds can be important - I use 0 for clean images, but increase it slightly for higher ISO images. You'll only see the effect of capture sharpening when viewing the image at 100%

Content / Creative Sharpening is something we want to be able to see when looking at the image as a whole - and since the resolution of most sensors is HEAPS more than the resolution of most monitors, it comes as no surprise that we want to have the contrast applied over an area that represents a bigger percentage of the image edges. The actual amount varies, but I typically use 4 pixels for portraiture. In terms of the amount, we need to be a whole lot LESS agressive than with capture sharpening, so often about 40% is right. Threshold is typically 0. It's important to note that this sharpening is appled to the full resolution image also.

Full resolution image are wonderful things, except when you want to display them online, and sometimes even when printing they're less than ideal - so often we need to downsize (or sometimes upsize), and in these situations, just chucking out "x" percent of the data on a linear basis doesn't always do the image justice; Imagine an image that has a sharpening halo that's 4 pixels wide in an image that's 4000 pixels wide. If we down-sample an image to 1000 pixels then the sharpening halo may then only be 1 pixel wide (which may be too small for our eyes to resolve) - so Bicubic Sharper will "fudge" the chucking away of info in such a way as to give priority to retaining more of these sharpening halos. On the other hand, Bicubic Soother will give LESS priority to retaining them if you're upsampling (because they may be visually too obvious). That's the theory anyway. Personally, I like to use normal Bicubic for down-sampling, and then reset the desired sharpness/look using output sharpening (I feel that I get a slightly better result).

Output sharpening (finally!) It can be a can of worms. If you're applying output sharpening to a down-sampled image for internet display then - because these images are usually displayed 1:1 (or best displayed 1:1) (ie 100%) then output sharpening of 0.3 pixel @ about 100% generally tends to do the trick nicely (it just gives it a final "clarity" that (sadly) oh so many people don't seem to have. Output sharpening for print is a different animal altogether though because there are additional variables of image PPI - printer DPI (and thus the translation between them by the printer driver) and print size. I print a lot of canvases around 44" wide and I just don't have to worry about output sharpening (the content/creative sharpening doesn't degrade becuase I'm not down-sampling and the image is big enough for folks to see the sharpening. On the other hand, an image printed post card size would require far more agressive sharpening for the increased edge contrast to be seen by the human eye (because the ability of our eyes to resolve something at a given distance is fixed, whereas the width of sharpening halos is a variable). So the first rule of output sharpening needs to be "the smaller the print, the more agressive you need to be with output sharpening). The other thing that enters the mix is the amount of information in the file that you're sending the printer; if it has a lot of information (say 720 ppi) then you may well get a different result than if you have essentially 180 ppi. So personally, I think it helps to down-sample your images to a consistant resolution prior to print so that you don't get any nasty surprises with consistency. (there are all sorts of claims as to what's needed; all I'll say is I use 180PPI and haven't seen any reasons to ever go higher.

In the interests of full disclosure ... I'm using PPI slightly incorrectly here, but I'm doing it deliberately because I don't want to confuse folks with DPI which is what the printer is actually spitting out. I don't think that a lot of folks realise that most printers spit out a constand DPI regardless of the resolution of the input file (eg my Epson 7800 always spits out prints at 1440 dpi even if I only feed in 10 ppi); it doesn't give you any more information in the image - it's just the way they work (the printer driver picks up the difference, which also affects sharpening - especially with high-frequency components).

So there's the background to it - not sure how much it helps, but all I can say is - if you're doing back-to-back tests, keep the resolution the same, but play around with different settings. An image that is well sharpened for output may well look severely over-sharpened on the screen, as screen and printer are vastly different devices.

As for actual numbers, sorry, can't help much with that as there are just too many variables. Looking at your initial setting though, I'd suggest changing to the ones I've mentioned above and see how you go (0.3/300/0 for capture sharpening, and 4/40/0 for content/creative).

Hope this helps!

[Boatman]

Colin:

Thank you for your thorough response. I think I'm on the right trail and perhaps worrying to much about details that don't matter. I am seeking the 2% improvement here.

I've got the general sharpening down with several techniques that work well. Ditto with the sharpening for the web. (Thank's to the excellent article posted a this site.) Where I'm confused, though again I'll state I think I'm worrying about it too much, is in the large-scale printer output settings. For an 11 x 14 or 11 x17 (which is what I frequently print), I usually have something more than 240ppi to work with. I've read that ideally for the Epson 1400 this should be upsampled to 360, but I can't see that it makes any difference. I've also found that if I've achieved good sharpening, viewed at 100% on screen, (minimal haloing, noise under control, etc.), I'm ready to print. I can add a touch of additional sharpening using Jeff Schewe's high-pass sharpening technique or I can add a little more in USM but the difference is imperceptable in the final print.

[Colin Southern]

Hi Homer,

Where I'm confused, though again I'll state I think I'm worrying about it too much, is in the large-scale printer output settings. For an 11 x 14 or 11 x17 (which is what I frequently print),

That's not large - I hardly ever print anything that small

I usually have something more than 240ppi to work with. I've read that ideally for the Epson 1400 this should be upsampled to 360, but I can't see that it makes any difference.

A lot of people have a lot of "good sounding theories", but at the end of the day, the printer driver is going to up-sample anyway, so who's to say that Photoshop will so a better job of that than the print driver? My suggestion is that it won't matter which one does the up-sampling, but you MAY get different results at different resolutions for given settings, so "eliminate the variables". For me that means usually printing at 180PPI - for you that may mean 240 PPI - but at least if it's consistent then you can eliminate one of the variables. But as you say, it's really only the final 2% you're chasing; unless it's a very small print, the bulk of your benefit will come from the content/creative sharpening (unless you want to optimise it for a particular viewing distance @ a given size). So many variables ...

I've also found that if I've achieved good sharpening, viewed at 100% on screen, (minimal haloing, noise under control, etc.), I'm ready to print.

It becomes more of a feel than anything else; you really can't make good content/creative sharpening evaluations at 100% (ie approx 100 ppi) for a print that's going to be a totally different size. Case in point: if you're printing small, you're going to need to have more agressive sharpening - and I'd be surprised if they WEREN'T visible at 100%. Noise is another issue; what's visible at 100% on a monitor will most likely NOT be visible in a print (for shadow noise anyway); prints just don't have enough dynamic range for it to show much.