| Home | Galleries | Articles | Reviews | Best Gear | Science | New | About | Contact |
by Roger N. Clark
This is an unsharp mask example in a series illustrating image sharpening methods.
The series:
Introduction to Sharpening
Unsharp Mask
Part 1
Part 2
Part 3.
Introduction
This page is dedicated to experiments in using unsharp mask (e.g. in Photoshop), and how it can be used to improve the apparent sharpness of images. Unsharp mask does not actually sharpen. Perceived sharpness in an image is a combination of the resolution of fine details in the image, the apparent size of those details as seen by the viewer, and the edge contrast (called accutance) in the image. Unsharp mask only addresses edge contrast. Even so it can be quite effective, and is simple and fast.
The Test
The Test target, Figure 1 was blurred using a Gaussian blur with a radius =2.0. This blur is similar to the blur that might be encountered in real images.
Figure 3 shows an attempt to restore the blurred image to the original test target state. It was done by converting from RGB mode to Lab, then running unsharp mask only on the lightness channel. Finally, the image was converted back to RGB. The advantage of using the lightness channel in Lab space is that colored highlights do not become white spots.
Try different combinations yourself. Save the images (I give permission to all to experiment with the images, just not for sale and profit), then try various ways to restore the blurred image to the original. The goal is to not have ringing artifacts (which appear as halos around objects). Figure 3 shows some ringing, especially around the black target. Try to not make the ringing any worse than in Figure 3. If you come up with a better solution, email it to me and I'll add it to this page (with credit to you if you wish).
The original test target, a 16-bit/channel tiff file, 1 megabyte, is here.
Figure 1. Original test target.
Figure 2. Test target blurred with Gaussian blur, radius=2.0.
Figure 3. Test target restored with unsharp mask, parameters indicated on figure.
Figure 4. Image restored using adaptive Richardson-Lucy deconvolution
using a 7x7 Gaussian point spread function, 200 iterations followed by
unsharp mask with radius=0.3, 150%, threshold=3. The result is clearly
better and with fewer artifacts than the result in Figure 3.
Discussion
Multiple pass sharpening is popular among photographers, and it is more effective than single pass. There is a scientific basis for multiple pass sharpening: the blur can be described in Fourier space as multiple frequencies. The unshapr mask using different radii is simply boosting those different frequencies that were reduced due to blur. However, unsharp mask is only enhancing edge contrast, which gives the impression of sharpness. It does not improve true resolution.
True resolution can be improved with image deconvolution methods, as shown in Figure 4. Deconvolution methods are iterative solutions and take some computer time. They tend to not be used by photographers because these methods are not in many popular image editing programs. But such methods are available in both commercial and open source image processing programs that actually cost less than the big systems like photoshop. Photoshop does not currently have multi-iteration deconvolution sharpening.
Most images in my galleries have had some unsharp mask applied. But a better method in my experience is deconvolution methods, and my standard workflow includes Apaptive Richardson-Lucy image deconvolution plus some modest unsharp masking. Deconvolutions methods can actually improve resolution as well as edge contrast, even correcting for some motion blur in an image. More examples of unsharp mask, image deconvolution and the two combined are illustrated in the rest of the sharpening series.
The series:
Introduction to Sharpening
Unsharp Mask
Part 1
Part 2
Part 3.
| Home | Galleries | Articles | Reviews | Best Gear | Science | New | About | Contact |
This page URL:
http://www.clarkvision.com/articles/unsharpmask
First published November, 2002.
Last updated February 22, 2014