Applying the Edge Mask
Now that the mask is completed, I load it into the image using Load Selection from the Select menu. (Note that this works with both 8-bit-per-channel and 16-bit-per-channel images, which means you can use all of Photoshop's tools to make selections in high-bit images. All too many people think that if you're working on a high-bit image, your selection options are limited to the marquee and the lasso. If that were the case, I wouldn't use high-bit images nearly as often as I do.)

With the selection loaded -- I always hide the "marching ants" by pressing Command-H (Mac) or Control-H (Windows) so that I can really see what's happening in the image -- I choose Unsharp Mask from the Sharpen submenu on the Filter menu. Since the filter is constrained by the mask, it's somewhat more forgiving in terms of the settings than it is when applied globally, but I always err on the side of caution: Remember, the goal is only to restore sharpness that was lost in the digitization process. In this case, I used a Radius of 0.4 pixels, an Amount of 140%, and Threshold set to zero to produce the result shown in Figure 8b. The original image is seen in Figure 8a, and a globally-sharpened version using the same settings is seen in Figure 8c.

Figure 8a: Unsharpened

Figure 8b: Sharpened as described above, using an edge mask and Unsharp Mask

Figure 8c: Original image after global sharpening, using the same Unsharp Mask settings used for Figure 8b.

The differences between the globally-sharpened version and the selectively-sharpened version are quite subtle at this point: If you look in the deep shadow areas, you can see slightly more noise in the globally-sharpened version. The real payoff comes when you sharpen the image for output.

Second Pass
After you've applied your selective sharpening, sharpen the entire image to get the best results with your given output device. For guidance on how you might best manage this, refer to my previous Out of Gamut column, "(Almost) Everything You Wanted to Know about Sharpening in Photoshop but Were Afraid to Ask."

The figures below show the image sharpened for output to an inkjet printer. Figure 9a shows the selectively-sharpened version after output sharpening: It appears slightly over sharpened on screen, but it's free of objectionable noise and grain and will make a good print. Figure 9b shows the globally-sharpened version after sharpening for output. Clouds of film grain seem to have exploded into existence, notably in the dark cloud at upper left, but also in the walls of the building. Most inkjet printers will use black ink to render the noisy parts, making them even more obvious, so this version will not yield a satisfying print.

Figure 9a: The optimum result, achieved by sharpening edge masking and then sharpening for output.

Figure 9b: A more typical (and less satisfying) results, achieved by using global sharpening and then sharpening for output.

It's true that this pre-sharpening technique involves a fair bit of work. But most of it can be automated: The only steps that require user intervention are making the grayscale rendition to get the best contrast, blurring the edge mask with Gaussian Blur, and making the Levels adjustment to the mask. If you're making multiple versions of an image for different outputs, a pre-sharpened master file is a good place to start: It lets you optimize your sharpening for the particular print process without worrying about image-content issues, because you've already addressed these. I process a lot of images, and I find that careful pre-sharpening saves time in the long run.

Reader Query
I'll conclude by answering a reader query posted in the VoxBox area of my previous column on sharpening. On Friday, January 26, 2001, Don Cooper wrote: "Would Bruce or anyone else please tell me why fractional radius values are used, since there is no smaller unit than the pixel, and we are told to work at output resolution?"

Filters such as Unsharp Mask, High Pass, and Guassian Blur use a mathematical method called a convolution kernel. This is typically a 3-by-3 or 5-by-5 matrix of numbers that represent a group of pixels. The pixel on which the filter is currently operating is always at the center of the matrix, so a 3-by-3 kernel works on the surrounding eight pixels, while a 5-by-5 kernel works on the surrounding 24 pixels.

The kernel changes the current pixel's value based on the values of all the pixels in the matrix. The Radius setting in filters like Unsharp Mask and Guassian Blur adjusts the relative weight given to the surrounding pixels: A smaller radius gives more weight to the current pixel, while larger radii give progressively more weight to the surrounding pixels. Photoshop doesn't carry out any operations on 0.4 of a pixel -- the pixel is indeed the fundamental particle in digital imaging -- it just uses the sub-pixel values to decide how much weight to give to the surrounding pixels.

Read more by Bruce Fraser.

Editor's Note: Click here for a step-by-step quick guide to Bruce Fraser's two-pass sharpening in Photoshop.