Unsharp Mask

[Brian Sims]

In the latest View Camera magazine there are two good articles on scanning large format film. The first, by Charles Cramer, strongly advocates not applying any sharpening (unsharp mask) tothe scan. In his work flow, he applies unsharp mask as a final step in photoshop. He says that the degree of sharpening is dependent on the target for your output. He says bigger files need more sharpening to achieve the same effect.

The second article, by Michael Mutmansky, advocates sharpening as part of the scan. he suggests that, if you don't know what your final output will be, apply just enough sharpening to just offset the softening inherent in the translation of analogue to digital. His overall goal is to get a scan that is pretty close to the image you want and then use Photoshop to just tweek the image.

I have been experimenting with both approaches and would be interested in comparing notes with others who are trying to (or who have) figured this out.

I'm using a Imacon 646, but previously used an Epson 4870. The Flexcolor software that comes with the Imacon has an additional unsharp mask control that Photoshop doesn't. With Flexcolor I can prevent any sharpening in the dark tones below a specify level. This eliminates the most obvious noise problems that result from sharpening. This seems to be one of the advantages of sharpening as part of scanning. Another is that my initial image in Photoshop is a little closer to my final goal and helps guide the rest of my adjustments. On the other hand, applying unsharp mask after setting all tonal adjustments makes sense. However, that process seems to take a lot more time because of the recommendation to save a master file, then flatten the image, then apply the unsharp mask. After doing so, I have often felt that further tonal adjustment were in order. That mean re-loading the master file and repeating the process.

Which approach are you all finding is most successful? How does resolution and print size affect your decision on how much sharpening to do (assuming you're working with the same subject matter)?

Thanks,
Brian

[Doug Dolde]

Listen to Charlie Cramer...he knows what he's talking about.

[Joseph O'Neil]

As a basic technique, I have always used my unsharp mask as my last step before saving an image, never as part of any scanning. Wait until you have a single speck of dust in your scan, apply unsharp mask to your scan and see what happens.

I use this technique for any and all sizes scans, outputs, file formats, etc.

joe

[Brian Ellis]

One of the difficulties with scanning IMHO is that seemingly equally knowledgeable people often make contradictory recommendations, which I suppose means there's no single way that's right for everyone. FWIW I prefer to use Photoshop for sharpening. Where and how much you sharpen depends on the content of the particular image. I almost never sharpen everything in the image by the same amount and there often are some areas (e.g. skies in landscapes) that I don't sharpen at all. So I use masks to sharpen, which allows me to select which areas get sharpened and by how much. That's relatively easy to do in Photoshop, it's impossible (AFAIK) in the scanning software I use (Silverfast and Vuescan). So I save all sharpening for Photoshop. However, I'm not a scanning expert, I'd be open to being persuaded that I should do some minimal amount of sharpening on the entire image in the scanning software.

I used to worry more about the content of the file that I sent from scanner to Photoshop before Photoshop allowed most editing to be done in 16 bit. Now I worry less about getting things as close to perfect as possible in the scanner. I set the black and white points and make major tonal and contrast adjustments in the scanner, everything else I save for Photoshop.

The best discussion of scanning I've seen is contained in Chapter 10 of the book Real World Photoshop. You might want to read it if you haven't already. Personally I would be very leery of any control that simply based sharpening on tone such as the Flexcolor program you mention. IMHO (and it's very humble, I don't claim to be an expert) that's far too crude a system. The noise problems you allude to can be cured without such a blunt tool by use of the Photoshop controls in USM.

[Paul Butzi]

One thing to be aware of is that the world of scanning, image adjustment, and image output is still advancing rapidly.

Sometimes there are fundamental advances - like manipulating images with a 16 bit depth instead of 8 bit depth.

In some cases, an advance means that things that were advocated before (like adjusting the image tonality at scan time, or up-rezzing in 10% increments) made sense before, but no longer make any sense at all and remain simply because that's the way people did it before and it takes a long time to change the conventional wisdom.

The good news is that experimentation is easy.

I scan images, adjust them, then sharpen them at output time. Often, I sharpen different regions of the print differently. I think that gives me the best results.

The good news is that it's easy to test. Scan the image twice - once sharpened, once not. Adjust both images as identically as possible. Perform the appropriate final output sharpening, and then compare the results.

And, please, post the results on the web, so we can benefit from the experiment, too.

[Michael Mutmansky]

Brian,

I think you misunderstood my article. I was not advocating any sharpening at the scan time. I only sharpen at the output stage, when the output size and resolution are known. This is the only point where it makes sense to sharpen. I was only writing about setting the endpoints at logical locations so that the scan does not lose information that may be desired in the final print. The main goal of the article was to get people to understand why the scanners produce a different result from silver gelatin paper.

USM is a destructive operation. It taked data in the file and alters it in a manner that is irreversible, and it should not be done in multiple steps, because each step of the way results in some of the original file integrity being compromised.

My point about scanning is that if you are not careful, you will lose the endpoints of the scan, and that can be detrimental to the quality of the final print that you can make. There are several points along the process that you have to make overall aestehetic decisions about the image, and the scanning step is one of these. I think that it should be treated as importantly as the rest, because it is the first opportunity in the digital realm to get the image 'right'. If you get it reasonably right in the first step, the rest will come relatively easily.

Be advised that the Imacon scanners have a 'dirty little secret' about sharpening. Even when you have them set to 'zero' sharpening, there is still some USM applied. There is a negative setting that can be applied to turn all USM off. This is part of the reason that the Imacon scanners appear to outperform other quality scanners. If I recall correctly, it is -200 or -120. I don't have an Imacon, so I have never committed that to memory.

Sharpening at the scan is logical for production scanning associated with a slide archive or other production-oriented application, as is ICE, but for the best possible results, you will do neither while scanning.

---Michael

[Frank Petronio]

I didn't read Michael's article but just from a practical matter, I think you need to apply at least some sharpening (I use the Smart Sharpening in CS2) to both my scans and digital captures in order to clean the edges up enough to know what you're working with. And generally to spot and dustbust. I don't try to sharpen all at once, and while it is destructive in an absolute sense, that's why you try to oversample and work in 16-bit.

From an absolute perfectionist POV, you probably should scan at exactly your output resolution, which in reality means working on test images and then re-scanning for that perfect final image, sized and cropped on the scanner. Because even downsampling introduces compromises as adjacent pixels must be averaged as they squeeze from 8 pixels to 1 pixel or whatever.

I am not a perfectionist, I just like to actually produce images in the real world. And like Bruce Fraser's excellent book suggests, it makes sense to create a solid, high res archive scan of a high bit image, with non-destructive adjustment layers and masks saved in layers, as the core to your workflow. Then you "save as" variations optimized for the web or various sized reproductions on different devices.

I've gotten digital images from other people for offset repro that look scary - over sharpened and harsh. They usually reproduce great. You usually have to take your sharpening to the point of making harsh halos before you'll see it look like "too much" for offset, at least at normal screens (133 -175). Inkjets actually show more, so they don't need as much sharpening. And web jpgs need the least and sometimes look better with a subtle blur.

[paulr]

I'd recommend that everyone read the (long) chapter on sharpening in Blatner and Fraser's Real World Photoshop series. They advocate a workflow-based approach to sharpening, which has given me very good results. They do advocate more than one sharpening pass at different times in the workflow, but in different ways and for different purposes. A very subtle sharpening ("capture sharpening") right after the scan, purely to regain MTF lost in the scanning process makes a big difference in retaining detail through subsequent operations. A final "output sharpeing," like what everyone else is describing, prepares the image for printing at final size.

I use layers to restrict sharpening in both cases to the midtone range of the image, so no pixels are blasted to black or white, and so grain and noise aren't emphasized.

Blatner and Fraser also talk about the possibility of an intermediate "creative sharpening" step, which you can employ if you want to sharpen up some part of the image that was fuzzy in the first place. This is more of a retouching technique, and I haven't done it except on other people's pictures. They suggest sharpening on a layer and using a brush to selectively apply it to the final image layer.

In any case, the value that seems to matter most is the sharpening radius of the final output sharpening. I put optical research to use and set the radius to 1/10mm, which emphasizes detail in the 5 lp/mm range. Done well the results are beautifully sharp, without any visible artifacts or digital crunchiness.

[Mike Kovacs]

I use photokit sharpener, which has tools for sharpening at the capture, and print stage, as well as intermediate "artistic" selective sharpening tools if appropriate, e.g. eyes and mouth of a portrait.

I believe its a product of the real world photoshop authors, and although I used to do a lot of the techniques manually, the results I get now are far superior. (in 35mm and MF - just getting started in LF)

[Brian Ellis]

I tried the Photokit sharpening software. The problem I had with it was that it works only in RGB if I remember correctly (I hope I'm not confusing it with one of the other USM programs I've tried). I scan b&w film in 16 bit at 4800 ppi. That produces very large files and converting them to 48 bit makes it basically impossible to work, my external hard drive just sits there forever with the red light on. If there's some way around this aspect of Photokit I'd be interested to know about it because I liked the program otherwise.