Idaho Airships, Inc.

Again we engage a contentious topic-one which requires a very deep knowledge of digital manipulation in order to master.

Because it enjoyed establishment within Photoshop, Unsharp Mask (USM) is the most frequently cited method of sharpening images. Unfortunately, USM is a relatively indiscriminate and therefore inappropriate technique-one that has been supplanted in the last few years by far superior algorithms. I describe USM as indiscriminate because it makes no attempt to identify what is and what is not an edge other than to simply measure the differences in values between pixels. For this reason, USM tends to significantly corrupt areas of noise regardless of their context (whether within a smooth gradient or situated upon a drastic edge), even with fairly attentive Threshold control, in the absence of controlled localization (which is not an intrinsic feature of USM).

I am unaware of any efficient, effective technique for preventing clipping forced by USM except for creating a Duplicate Layer (Normal Mode), applying the USM, the Splitting the Sliders to confine the USM to an acceptable Luminance levels. Simply limiting the Output Level also affects areas that are marginal independently of the USM.

This is not a basic tutorial on Unsharp Mask. If you need a good primer on the matter, please check the following sites:

Photoshop (CS3) Help

Cambridge in Colour Image Sharpness Essay

Cambridge in Colour USM Tutorial

To this day there are still widespread misconceptions about USM. Many of those misconceptions are either born of or exacerbated by the necessity of viewing the image on a monitor during USM applications, which is an improper technique at its core. The way an image appears on a monitor (even at View>Print Size) and the way it appears in print are profoundly distinguished.

The confusion about USM is so well established that there are still very erudite discussions about whether USM is a convolution filter or not. I’m convinced that it is a linear convolution vice the matrix-style convolution available in Filter>Other>Custom (CS3).

But all of this is moot, or should be, in the forensic imaging realm. Digital sharpening is inarguably a corruption of image virtue for the sake of aesthetics. Forensic images shouldn’t be striving to be visually pleasant-they should be striving to be accurate on all fronts (an admittedly impossible goal). Sharpening is prima facie inappropriate for forensic images excepting when used as an investigative preparation.

Another issue about sharpening that is generally ignored outside of the more sophisticated algorithms is that sharpening should be adjusted for viewing distance. Just as viewing distance (or more correctly, viewing from the actual center of perspective) is the single most important factor in image (viewing) accuracy, the spatial character of sharpening should conform to the linear enlargement model used to determine that center of perspective. Various formulae for the application of USM have been suggested, some of which are very impressive, but they still rely upon the indiscriminate fundamental nature of the filter.

Sharpening in other than the working gamut is clever and does provide some theoretical and actual advantages, but it also carries the USM baggage. Confining USM to a single or two of three Channels isn’t necessarily any more elegant…it amounts to simply avoiding potholes. All of these workaround methodologies are simply steps in the development of the ultimate process predicated on minimizing the penalties of existing techniques.

So, what is the ultimate process?

For monitor or device (cel phones, PDA, et. al.) display, I simply use the Filters>Sharpen>Sharpen function, and Edit>Fade Sharpen to adjust the effect down if desired. Keep in mind that any species of sharpening frustrates .jpg encoding (the effect is more pronounced at more aggressive compression levels). This is typically only an issue when preparing images for low-bandwidth monitor or device display, but the effects of eschewing sharpening or even introducing a bit of blur or are significant and can present a remarkable advantage in select instances (e.g. Flash Lite).

72dpi, Gaussian blur of .3, .jpg 7/12, file size = 59K

72dpi, no blur, no sharpening, .jpg 7/12, file size = 60K

72dpi, Filter>Sharpen>Sharpen, .jpg 7/12, file size = 71K

For my non-forensic print workflow I use the S-Spline algorithm commercially available in PhotoZoom Pro. I was very diligent in selecting such an important element of my non-forensic workflow. PhotoZoom Pro addresses all of the resizing, resampling, and sharpening issues a professional shooter is likely to encounter, and I recommend it without hesitation.

Once an image is converted from RAW, it should not be resized without the S-Spline application, and that resizing should be the very last preparation step prior to printing. Do not sharpen an image in camera (technically a misapplication of the Latin legal term, but so very appropriate in other ways).

PhotoZoom works both as a standalone application and within PS as a plugin.

There are some very inventive alternatives to a global application of USM, whether in the working space on all Channels or selectively per Channel. One is to simply duplicate your image, apply USM at the desired potency, then Clone the actual edges from the duplicate to the parent image. Another is to duplicate the Background Layer and apply the USM there, then Mask out areas where the USM is undesired. Toward that end, some rather sophisticated Masking techniques can come in handy. Or they can drive you to drink. Always keep in mind that well-developed Masking techniques involve very few steps…please don’t be intimidated by what appear to be extremely complex results.

Fully prepared architectural aerial of a Salt Lake City, Utah subject.

L