The following are some general considerations for all sharpening techniques. These are "rule of thumb" considerations. As with all "rule of thumb" statements, there are always exceptions, but these rules do serve as a good starting point.
USM is a tool for sharpening digital images. USM identifies areas of high contrast between pixels. It then increases that contrast. While USM is a very popular tool, it is sometimes applied in an inexact manner. In order to produce the best results, it is necessary to understand the USM settings. Therefore, the three settings will be covered first. Once the settings have been covered, a general procedure for determining the best settings for a print will be introduced.
USM is opened by selecting Filter/Sharpen/Unsharp Mask. Figure 1 shows the USM tool. USM has three settings: Amount, radius, and threshold.
The amount setting controls the intensity of the sharpening. The higher the setting (from 1% to 500%) the more intense the sharpening. Intensity is defined as how much the contrast between adjacent pixels is increased. In other words, the amount setting determines how light USM makes the lighter pixels and how dark it makes the darker pixels.
Figures 2 through 4 (shown at 50% view) show an image at various amount settings (the radius and threshold settings were held constant). Figure 2 shows the image with no sharpening. Clearly, the image is too soft for printing. Figure 3 shows the image with a sharpening amount of 75%. The wood, plaster, and bottles have picked up more texture. Figure 4 has the amount increased to 160%. At this amount, the image is showing signs of oversharpening. With the amount increased to the maximum of 500%, Figure 5 shows severe oversharpening. Notice how increasing the sharpening too much has actually caused some loss in detail in the wood, rather than enhancing it.
The important point here is that more is not necessarily better when it comes to sharpening. The idea is to find the optimal sharpening settings -- not to increase them to the maximum. Starting from an unsharpened image, increasing the amount of sharpening will bring out image detail, up to a point. After that point, additional increases in the amount will actually degrade detail. The results will be seen as the granular look of oversharpening.
The size of the final print also needs to be kept in mind. The larger the print, the more the sharpening amount can be increased.
The radius determines how wide of an area, around a particular edge, will be sharpened. A low radius means that the sharpening will be applied right next to the edge only. A larger radius applies the sharpening further out from the edge.
It was mentioned, in Part I of this series, that USM creates a copy of the original layer, blurs the copy, sandwiches the two copies, and calculates the difference in tonal values between the original and blurred images. What actually happens is that the radius setting determines how much blur is used when blurring the duplicated copy of the image. It is important to understand that a radius setting of one pixel does not mean that the sharpening is applied across a one pixel width (it would actually be applied across a width wider than one pixel).
Figures 6 through 9 (shown at 50% view) show an image at various radius settings (the amount and threshold settings were held constant). Figure 6 shows the image with no sharpening. Figure 7 shows the image with a sharpening radius of 1.0. There is some improvement in the image sharpness. However, it is slight and still not good enough for printing. Increasing the radius to 2.7 (Figure 8) causes the grain of the wood and the chipping paint to really stand out. This is probably a good radius setting for this image for inkjet printing purposes. Figure 9 shows the image at a radius setting of 7.0. At this setting, the image is starting to show signs of oversharpening (notice the degraded grain in the board at the left side of the image). This is too large of a radius for printing applications with this image.
The threshold determines how much tonal difference must exist between two pixels before sharpening is applied (on a 0 -- 255 scale). A setting of 0 means that almost any tonal difference will be sharpened. High threshold settings will result in areas being sharpened only if there are large tonal differences between pixels. For example, a threshold setting of 5 will result in sharpening being applied only when adjacent pixels have a tonal difference of 5 levels or more. Threshold is used to prevent USM from sharpening areas that have little contrast. It is also useful for preventing USM from sharpening noise.
Figures 10 through 13 (shown at 50% view) show an image at various threshold settings (the amount and radius settings were held constant). Figure 10 shows the image with no sharpening. The sharpening is first applied in Figure 11. It is valuable to compare Figures 10 and 11. In Figure 10, there is little detail in the bird's feathers. After sharpening, the very fine detail of the feathers becomes visible in Figure 11. The texture of the feathers has become visible just behind the bird's eye as well as ever so slightly on its chest. This is a very good thing. On the other hand, the sharpening has also brought out noise in the water that forms the background. This is not a good thing. One way to deal with this noise is to increase the threshold. Figure 12 shows the image with the threshold increased to 2. The noise has been somewhat attenuated. However, the detail in the feathers has also been decreased. Increasing the threshold to 10 has almost eliminated the noise. It has also pretty much completely eliminated the feather detail.
Okay, so this setting does this, and that setting does that. We still don't know how to determine the optimal settings for an image. Now, as mentioned at the beginning of this article, nobody can give you an absolute answer to your sharpening tasks (e.g., put the amount to this setting, the radius to this, and the threshold to this). Rather, we will review a step-by-step procedure for sharpening images.
Step One -- Set the Sharpening Radius: Perhaps the most important key to getting the sharpening right is to find the correct radius. Thus, the radius should be set first. Once the correct radius is found, the amount and threshold are much easier to determine.
The optimal radius setting is affected by a number of factors. For starters, the setting is affected by the capture device, output device (e.g., continuous tone vs. inkjet printer), and image size. However, the reality is that most photographers tend to capture primarily with one or two devices, print primarily on one or two types of printers, and use only a few sizes of paper. For instance, most of my images are captured with my favorite digital camera and are printed on the same inkjet printer as either 8" x 10" or 13" x 19" prints. This greatly simplifies the process. Once a photographer has built up a bit of experience with her favorite equipment and print sizes, those factors pretty much become a constant and are no longer an issue. The photographer can then concentrate on the last two major factors that impact the optimal radius setting.
The last two factors are image content and image resolution. With respect to image content, the important issue is the size of the image detail. The finer the image detail, the smaller the radius must be to render that detail. The coarser the detail, the larger the radius required. Thus, a close up of a peacock with the fine detail of the colored feathers will call for a smaller radius while a shot of a baby's smooth face will necessitate the use of a much larger radius. However, image resolution also plays a part. Fine detail from a low resolution camera will be spread over a small number of pixels. Thus, a small radius would be required. However, that exact same detail will cover many pixels if shot with a very high resolution camera. In this case, a much larger radius would be required to properly render that detail.
Considering all of these factors, how does one find the sweet spot for the radius? The following procedure generally produces good results. Set the radius at an initial value between 1 and 3, the amount at a fairly high value (300% -- 500%), and the threshold at 0. The high amount and low threshold will likely produce greatly exaggerated sharpening. This is okay for now because it allows the photographer to see what impact different radius settings have on the image. The goal now is to move the radius slider around while observing the image at the 100% view. We want to move the radius to the point where the image detail begins to degrade or a noticeable halo appears around the edges. Then, the radius setting should be backed down a bit. This will be the initial radius setting (I say initial because it may be changed later based on the results from a test print).
Experience also plays a part. Over time, a photographer will develop a feel for the best radius setting for a particular set of circumstances as well as for how that radius setting will look on the monitor. For any new image, it is then just a matter of playing with the radius setting until that look appears on the monitor.
Once the photographer is satisfied with the initial radius setting, it is time to move on to the second step.
Step Two -- Set the Sharpening Amount: The next step is to adjust the sharpening amount. For inkjet printing, I recommend that this step be performed at the 50% view (if you are using other output devices, you will need to find out what works best for you). Now, the amount should be adjusted until the image looks a bit oversharpened. Keep in mind that the larger the final image, the more oversharpened the image should look.
Step Three -- Set the Sharpening Threshold: The threshold is the last setting to be determined. Again, for inkjet printing, I recommend that this step be performed at the 50% view. The threshold should now be moved to the point that protects the smooth areas and noise from being oversharpened. This will likely be a fairly small value (I often set the threshold at 0; my digital camera has very low noise, and I prefer to protect the smooth areas in other ways).
It is now time to make a test print. The test print will provide indications for adjustments if necessary.
These settings are not independent of each other. For instance, if the radius is increased, the amount will need to be decreased or oversharpening will occur. Also, higher levels of amount can be accompanied by higher levels of threshold. The possible number of combinations of the three settings is seemingly endless. So, to prevent you from driving yourself crazy, focus on getting the radius setting properly matched to the image detail and resolution first. Once that is set, determine the amount and threshold settings that provide the look that you want for the radius that has already been determined. In other words, follow the three step procedure.
Figures 14 through 20 show an image going through the three step process. Figure 14 (shown at 50% View) shows the unsharpened image. An analysis of the image shows that this image has a lot of very fine detail. The camera that produced the image is a high quality eight megapixel camera. So, the resolution is decent. Despite the decent resolution, some of the detail is so fine that I suspect that the image will require a fairly small radius.
The monitor is now set at 100% view, so that the effect of the changes on the radius value can be better seen, and USM is launched. It is decided to start off with a radius of 1.0. The sharpening is set to a very high value of 400 while the threshold is set to zero. These somewhat extreme amount and threshold settings will exaggerate the effect of the radius, which will make it easier to examine what happens as the radius value is changed. Figures 15 through 17 show the image at different radius settings.
While the initial radius setting of 1.0 is probably a good first guess, a little experimentation is in order to determine how changes in the radius will affect the image. Thus, the slider is slowly moved to the right to increase the radius. The result is that the detail in the image quickly degrades. At a radius of 2.0 (see Figure 15), the detail is heavily damaged. Obviously, a larger radius is not acceptable (exactly what I expected). It is now time to decrease the radius. At a value of 0.8, suddenly detail jumps out that I didn't even know existed -- almost like magic (see figure 16). Dropping the radius even further produces no more detail but causes the image to grow soft (see Figure 17). It is important to compare Figures 16 and 17. While Figure 16 appears harsher than Figure 17, it contains all of the detail that is in Figure 17. The harsh look of Figure 17 is okay at this point because it is due to an artificially high amount setting (that will be adjusted later). Thus, increasing the radius up to 0.8 makes the detail stand out without degrading it. Further increases in the radius cause the detail to degrade. It appears that a radius of 0.8 is best for this image.
The monitor is now set at 50% view for the setting of the amount and threshold. Figures 18 through 20 (shown at 50% view) show the image during the final steps of the sharpening process. Figure 18 again shows the unsharpened image. Figure 19 shows the image at the current settings of radius 0.8, amount 400%, and threshold 0.
Figure 19 clearly shows that the amount setting of 400% is way too high. Some experimentation leads to a decision to set the amount at 225%. Finally, there are no areas with noise problems and no smooth areas with which to contend; there appears to be no reason to move the threshold value from its setting of 0. Figure 20 shows the image at its final settings of radius 0.8, amount 225%, and threshold 0. This image appears a bit oversharpened on the monitor, which is what is expected for an image that will be printed on an inkjet printer
It is now time for a test print. The test print will reveal whether the sharpening is as desired. If not, it will indicate what changes need to be made.
Figures 21 through 27 show another image going through the three step process. Figure 21 (shown at 50% View) shows the unsharpened image. It can quickly be seen that this is quite a different example than the last one. This image has no fine detail. Instead, it is a "low frequency" image with smooth detail. In addition, the camera that produced the image is also a high quality eight megapixel camera. So, again, the resolution is decent. Based on this information, I suspect that the image will require a much larger radius than the Bristlecone image.
The monitor is now set at 100% view so that the effect of the changes on the radius value can be better viewed, and USM is launched. It is decided to start with a radius of 2.5. The sharpening is set to a value of 350 while the threshold is set to zero. Figures 22 through 24 show the image at different radius settings.
As in the first case, a little experimentation is in order to determine how changes in the radius will affect the image. It is decided to start by increasing the radius. I quickly discover a rather disconcerting problem. Because the surface of the berries is so smooth and lacking in detail, I can not really tell when the radius has become too large. It can be seen that the berry surface becomes rougher as the radius is increased, but there is no clear-cut point where it can be determined that further increases in radius degrade the detail (as there was in the Bristlecone example). Not too worry. There is an easy solution. An examination of the image indicates that the middle water drop on the berry has some detail in it. It can be seen in Figure 24 that the detail is present in the water drop at a radius setting of 1.0. Conversely, it can be seen in Figure 22 that the finer detail has been destroyed at a radius setting of 5.0. Somewhere between a radius of 1.0 and 5.0 is the point where the radius becomes optimized for this image. Some playing around demonstrates that increasing the radius up to 2.3 makes the detail stand out without degrading it. Further increases degrade the detail. Thus, the radius is set at 2.3 (see Figure 23).
The monitor is now set at 50% view for the setting of the amount and threshold. Figures 25 through 27 (shown at 50% view) show the image during the final steps of the sharpening process. Figure 25 shows the unsharpened image. Figure 26 shows the image at the current settings of radius 2.3, amount 350%, and threshold 0.
Not surprisingly, Figure 26 clearly shows that the amount setting of 350% is too high. Some further experimentation results in the amount being set to 150%. It now becomes important to protect the smooth areas from being oversharpened. I notice that the USM is trying to sharpen the surface of the berries. This produces a strange look. Moving the threshold setting to 3 removes this undesirable sharpening without a significant degradation of image detail. Figure 27 shows the image at its final settings of radius 2.3, amount 150%, and threshold 3.
As always, a test print will be made and examined. If necessary, the sharpening settings will be adjusted based on an analysis of the print.
Now that you have spent so much time in order to understand all of this detailed information, you probably expect me to pat you on the back and wish you well in your sharpening endeavors. Fat chance! Remember what it said in Part I, "In short, this article is a thinking person's guide to sharpening". It doesn't take much thinking to discover that this approach has some serious problems. This is a one pass approach. It applied USM equally to the entire image. The problem is, the entire image did not have equal needs for sharpening. To better understand this, Figure 3 has had some arrows added to point out problems. This image has been reposted as Figure 28 (shown at 50% view).
Arrow 1 shows a window pane that has been sharpened just about the right amount. The weathered wood grain clearly shows without any oversharpening. However, arrow 2 shows another section of the window that appears less weathered. This wood did not respond as much to the sharpening; it is somewhat undersharpened. Arrow 3 shows a section of the cement like material that was used to hold the bottles in place. This part of the wall is beginning to show signs of oversharpening. So, this image is properly sharpened, undersharpened, and over sharpened -- all at the same time.
The problems with one pass sharpening don't stop there. In images with noticeable noise, sharpening has a tendency to sharpen that noise, which makes it stand out and look much worse than before sharpening. Of course, you can always increase the threshold to decrease the noise, but only at the cost of losing the fine detail in the image (as was seen in Figure 13).
The sharpening requirements of the image content often conflict with the sharpening requirements of the output device. For example, you may have an image with very fine detail. You want to protect that delicate detail from oversharpening, so you intend to keep the sharpening levels fairly moderate. On the other hand, you want to make a large print, and large prints require higher levels of sharpening.
This process of sharpening can cause color fringing (color halos appear around the sharpened edges).
Sharpening is destructive; it degrades the quality of the image (this is the case for all sharpening methods). However, the damage caused by this method of sharpening is permanent. You can not go back on a later day and undo the sharpening in order to eliminate the sharpening damage or to resharpen for a different output device or size.
What's a poor photographer to do? I suggest that you read the subsequent parts of this series. All of these issues will be addressed.