Sharpening Tutorial - Part 1

Submitted by Bryan Wolf on Wed, 01/05/2011 - 5:03pm

Introduction

Over in the "Post your medium format photos!" thread at Pentax Forums, someone asked about highpass sharpening and how it differs from other sharpening methods.  I don’t claim to be an expert on this subject and the majority of the information below is more thoroughly covered in “Real World Image Sharpening” by the late Bruce Fraser. A book full of more information and examples than I could ever begin to put down here. I fully recommend picking it up, and there is a newer 2nd edition for you Lighroom users too!

This tutorial will be a 3 part series:

• Part 1: General Overview(Now online)
  I’ll do my best to explain, with examples, three of the more common forms of sharpening as I understand and use them. I will not go into much detail of what all the settings do or which values to use, as that will be thoroughly covered in part 2.

• Part 2: Workflow (Now online)
  I’ll touch on masking noise/grain when optimizing sharpening, adjusting local contrast, and I’ll briefly touch on Lightroom’s very good “Detail” pane. I will also do my best to debunk a common myth that highpass sharpening is in someway better for film captures, and show that in reality, these different sharpening methods are more alike than different.

• Part 3: Application (...Coming soon!)
  I’ll show the effects of these sharpening methods on what matters most, the prints. As an added bonus, I’ll also be including a set of photoshop actions that will help you more easily apply the techniques covered in this tutorial.
   

Disclaimer

First the obvious... Sharpening can in no way bring back lost resolution from lens or motion blur. However, it can restore some snap to images that lack it, particularly in a print where loss of sharpness is to be expected(from ink/photons bleeding together).

Second, because this is a general overview, I have oversharpened all images so we can better study the effects. I do not reccomend such high settings for general use, and suggest that you always use as little sharpening as needed to achieve the goal; i.e. resotore lost sharpness ,but keep your edits invisible.

The Target

I’ll be using the following image for the initial explanations and tests; a 4”x1” image at 360PPI reduced to 960px wide for display here(click any image to enlarge):

Figure 1 - Base sharpness test image

Non-Destructive Sharpening

I do all my sharpening on a separate layer(s). Sharpening on a separate layer gives you the option of  fine tuning the layer opacity after the filter has been applied. Also, as an added bonus, this method is completely non-destructive to your original capture! When using unsharp mask or smart sharpen, I set the sharpening layer to "Luminosity" blend mode for color images, and "Normal" for grayscale images. To do this simply duplicate your base layer, change your blend mode, and apply your sharpening. Alternatively, for muti-layer images, you can use "Merge Visible with copy" by holding the "alt" key and selecting "Merge Visible" from the layer palette’s menu; illustrated in Figure 2 below. Likewise, I use a separate layer for highpass sharpening, but with a different layer blending mode. I’ll detail this further in the Highpass section of the tutorial.

Figure 2 - Creating a sharpening layer

Multi-Pass Sharpening - a brief overview

I'm a firm believer in multi-pass sharpening. I feel it offers the most control and in my experience, the best result. It may not be the fastest workflow, but results are what's important to me and I'm sure many of you too. For those of you unfamiliar with the multi-pass workflow, I'll explain.

• Capture Sharpening
  Every captured image can benefit from some degreee of initial sharpening to retore loses from the capture process. This is true, be it direct capture via a digital camera, or a scan of film.  We want to restore most of the lost sharpness and nothing more. With this in mind, I always do an initial round of sharpening at a very low radius and amount, typically never exceeding 1 pixel and 50-100% for the amount.

• Content Sharpening
  The second round of sharpening is for content optimization. A portrait might get less than a landscape for example. Like the first step, I never use more than  the minimum amount needed to achieve the result I'm after, and I avoid visible halos like the plague.  I don't want to see halos, simply because an image may be printed at any number of sizes, and If the image needs any interpolation for larger prints, then visible halos will also become bigger.

• Output Sharpening
  The third round of sharpening is completed only after the image has been resized to it's final print dimensions and resolution. I print with an Epson printer, so that is always 360 or 720PPI at whatever dimensions the print may be(i.e. 5"x7" 360PPI). Unlike the other rounds of sharpening this value does not usually differ based on content or capture. A printer will always loose some sharpness and we can expect it to always be the same amount from print to print(assuming the same paper). Based on some science about human vision,we can roughly estimate the size, in pixels ,that halos will become visible. Assuming a normal viewing distance(12"-24"), that value is around  3/100 of an inch. So in a 300PPI file we might try and shoot for sharpening halos of around 3 pixels. 

The Candidates

Basic Unsharp Mask

Figure 3 below, has been intentially oversharpened to better illustrate on screen what’s happening. As we can see, dark and light halos have been added to the edge, in effect creating edge contrast. We also encounter our first problem which is better illustrated in the 400% crop(Figure 4).

Figure 3 - Amount: 250, Radius: 3, Threshhold: 0

Figure 4 - Amount: 250, Radius: 3, Threshhold: 0, 400% crop

In the above crop we can see that the sharpening halos have caused highlight and shadow edges to clip to pure black and white. What we really want, is a way to restrict sharpening to the midtones where contrast is most needed and away from areas that are already high in contrast. Photoshop’s layer blending options allow us to do just that.

Layer Blending Options

Access the "Blending Options" by double or right clicking the sharpening layer. The "Blend If" section, marked in orange below(Figure 5), is the area we need to adjust. Basically, "blend if" allows you to select which shades of gray on the sharpening layer will affect which shades of gray on the layer below it. You do this by adjusting the sliders to the values you want protected. By holding down the "alt" key you can separate the handles to create a falloff and therefore, smoother transitions. I’ll use the settings in figure  6 below for all of the remaining illustrations:

Figure 5 - Double click sharpening layer for blending options

Figure 6 - Blending Options

Below(figure 7) is the result. Compared to Figure 3, which uses the exact same unsharp mask settings, we can see a much smoother and more subtle effect without any of the harsh clipping of extreme values.

Figure 7 - Amount: 250, Radius: 3, Threshhold: 0, w Layer Blending Options

Smart Sharpen

If you compare the results below(Figures 8 & 9) to Figure 7 above, it should be obvious that with the same settings, smart sharpen using the "Lens Blur" algorithm will give a more subtle effect . It achieves this by being both lower in contrast and adding dark and light halos to either side of the edge. If "Gaussian Blur" is selected in the Smart Sharpen dialog, rather than ‘Lens Blur’ as above, the halos become larger, roughly equivalent in size to unsharp mask’s, but they are lower in contrast by comparison.

 
Figure 8 - Basic, Amount: 250, Radius:3, Lens blur, More Accurate

Figure 9 - Basic, Amount: 250, Radius:3, Lens blur, More Accurate, 400% Crop

High Pass

And now to the original question! Highpass sharpening adds contrast to high frequency detail(i.e. edges)using the "Overlay" blend modes combined with Photoshop’s highpass filter.

First I’ll take a second to explain why we use the overlay blend modes. Overlay increases contrast by combining the "Multiply" and "Screen" blend modes  and the very special value of 50% gray. 50% gray is special because when used with the "Overlay" blend mode, nothing happens... However, any value lighter than 50% has the effect of lightening the layers below it. While any value darker than 50%, darkens the layers below it.

Figure 10 - 50% gray overlay on left

As an aside, this can be a very handy way to do a simple, non-destructive, dodge and burn layer. Create a new layer and edit>fill with 50% Gray. Then set that layer to the "Overlay" blend mode and use the "Dodge" and "Burn" tools to, well... dodge and burn. : )

Softlight works very similarly but with less contrast, as does "Hardlight", but with more contrast. In figure 11 below, I’ve taken a base 75% gray layer and  placed a gradient layer of 0-255  on top of it. I then changed each gradient layer to the labeled blend mode.

Figure 11 - 75% gray base with Softlight, Overlay, Hardlight blend modes

Figure 12 - The layer setup to make Figure 11

So what’s this got to do with highpass sharpening? When you apply the highpass filter, smooth areas of an image get turned to that magic 50% gray. Darker edges get a darker than 50% halo, and lighter edges get a lighter halo. For example, Figure 13 has had the highpass filter applied with a radius of 3 pixels and was then resized for display here. Remember, 3 pixels is a larger value than I would normaly reccomend, but it makes it easier to spot the effects.

Figure 13 - Highpass applied to a pictorial image with a radius of 3

With our knowledge of how the overlay blend modes work, we can begin to visualize what effects this image would have on underlying layers. To illustrate further, the figures below, show the effect of  a 3 pixel highpass combined with the Softlight, Overlay, and Hardlight blend modes.

Figure 14- Highpass with a radius of 3

You may be saying to yourself, "I can’t see the difference". I’d agree, the differences are very subtle here. However, I can tell you that when using highpass sharpening on pictorial images, the difference becomes much more obvious. See figure 15 below. iI have taken out the gradients and replaced the base layer with 66% gray, effectively showing just the halos. The differences should be much more obvious now, and the results are similar to what we saw in Figure 11.

Figure 15 - Highpass radius of 3 with just the edges

Conclusion

So why would you choose one over the other? Well there is no real answer to this, I can only give my opinion and it’s probably best for you to decide after playing with all 3 in your workflow. I’m just some guy on the Internet and you are the better judge of what you want or need. With that said, these are the conclusions I’ve came to.

Unsharp Mask can add large amounts of high contrast edge effects. Therefore, it may be a good candidate for overly soft images or where you want a bitingly crisp result. It also makes a good candidate for a final application of print sharpening, where it’s high contrast artifacts will not be magnified by further image editing.

I generally prefer the more subtle effect of Smart Sharpen for initial capture sharpening and feel that the results add more than enough snap to most of my scanned images, but not so much that they’ll artifact badly with further processing.

I like highpass sharpening when optimizing sharpness for content. For example, on a portrait I may use the "Softlight" blending mode with a low opacity, and a landscape might get "Hardlight" with a higher opacity. The control in fine tuning you get with high pass sharpening is it’s main seller in my opinion. Some on the Internet have suggested that  the "High Pass" filter is better at separating film grain/noise from edges. In my testing, I have found the differences to be so subtle as to be a non-issue. I believe there are far better ways to restrict sharpening to edges and I'll detail these masking techniques further in part 2 of this tutorial.

So that’s it! I know this was more of a general overview than a tutorial, but I wanted to create a base for the workflow techniques I’ll introduce in part 2. Check back soon for the shocking conclusions! : P ...Oh, by the way, feel free to leave me a comment! This is my first tutorial, so it may be a little rough around the edges and any constructive criticism you have to offer would be greatly appreciated!

‘til next time,

Bryan

• Login to post comments