By Crawford Hart.
Part 1 of this series provided an overview of some of the generally accepted approaches to the problem of achieving perfect skin. The techniques themselves aren't glamorous, though the results can be. They tend, rather, to be boring, repetitive, simple procedures, tasks in which the operator must do most of the work. Grunt work, in other words.
Part 2 (the one you're reading now) and Part 3 (to come) will shift over to those effects in which Photoshop does the work and we just add up numbers and push buttons, though a little bit of delving into the esoterica of Photoshop, how it works under the hood, will be called for. But the payoff is definitely worth it. The ideas behind the process discussed here certainly did not originate with me; they can be found on other forums all over the web. I've provided my own particular slant, as I would expect you to do, once you're comfortable with the basic ideas.
...to here, without lifting a clone tool or healing brush, without dodging or burning or patching or anything else. Now, it must be said, the image we are starting with comes in pretty good condition, both in terms of the photography, and in terms of the skin quality. No scars, ridges, bumps or gross discolorations. So, of course, we're going to degrade things a bit before embarking on the restoration.
This shows the result of some channel blends that I've been playing around with, splitting up the R, G, and B channels and blending them back into the image in various modes to create more dramatic contrast. These blends and their possibilities will be covered in Part III, but, as you can see, including the blue plate to add weight also brings out imperfections that were mostly yellow discolorations in the original. An ugly mess. if we could just get rid of the blotches, the gain in contrast would definitely be worth it. It's out of the question to try to dodge and burn them away, and the healing brush would just stir up the soup. In fact, it's such a disaster that the idea of even trying such a move probably would never come up in a normal work flow. Better to just rely on curves for contrast, though I've found that they don't do the job quite as well.
And of course, blurring is a rigid taboo. Any blur that could get rid of the blotches would certainly lose any useful detail. Wouldn't it? Turns out, the answer is "Not necessarily."
THE BASIC IDEA The theory sound simple enough: blur the big stuff and leave the small stuff alone. We want to lose the blotches but keep the pores and other fine details that make skin texture believable, and, the lack of which, conjours up the dreaded judgment "AIRBRUSHED!!!" But if it sounds simple, it also sounds impossible, until we actually explore the relationship between blurring and sharpening in Photoshop, specifically between the Gaussian Blur filter and the High-Pass filter. Rather than get into an elaborate discussion that is ultimately beside the point, try this yourself with any image you wish. Or you can just watch me as I play around. NOTE: If you chose to pull the example images posted here, your numbers will be different than mine. I worked with an original Hi-res image; these are scaled down for the purposes of illustrating the steps.)
BLURRING AND SHARPENING Start by duplicating an image twice into new layers. Leave the original alone. We'll use it for comparison. To the first copy apply a Gaussian Blur such as you see here. The radius will vary, depending on the image size and resolution. The goal is a broad blur that wipes out all fine detail, leaving only areas of color. In this case I used a radius of 25, but the numbers aren't important. Getting rid of the details is the point.
While the Gaussian Blur filter is well-known, the High Pass Filter is a little stranger, less user friendly and not so easily understood. It is often used for various sharpening moves. Depending on the radius, it reveals fine detail against a 50% gray background. The larger the radius, the more detail is exposed. When used as a sharpening tool, the idea has been to combine the resulting layer with the underlying image in one of the modes that both lighten and darken, based on the value of the pixels on the top layer. The usual modes are Soft Light, Overlay and Hard Light, modes for which 50% gray has no effect and so only work on the level of detail revealed.
Go now to the top layer in your file and choose Filters>Other>High Pass.
Like Gaussian Blur, there is only one value to set, and it too is a radius. Use the same value that you used for your Gaussian Blur layer. It will look more or less like this.
Note how each layer contains precisely the information that the other is lacking. It might have been more helpful had Adobe extended the Audio metaphor of the High-Pass filter and called the Gaussian Blur filter a Low Pass filter.
Note how effectively the High-Pass layer can restore detail to the blurred layer underneath, when used in the modes previously mentioned. All of these are at 100% opacity. If you turn off the blurred layer, so that the High-pass layer is blending into the original image, you will see how the filter can be used as a sharpening method. It's effective, but not nearly as effective as it will be once you understand what is really happening and can exploit the effect to its fullest potential.
TWO SIDES OF THE SAME COIN
Turn the blurred layer back on
Set the blending mode of the high pass layer to Linear Light, and the opacity to 50%. (The math isn't relevant to using the modes correctly; it's enough to know that Linear Light has a much stronger blending effect than any of the other modes. It is the same as using Additive in Calculations or Apply Image.)
With the exception of the very lightest highlights, which have gotten lost in translation, we have not only restored detail, we have virtually returned the image back to its original state. Toggling back and forth from the bottom layer alone to the two layers on top will quickly confirm this.
You can experiment with the opacity of the High Pass Layer: raising it above 50% sharpens the image, lowering it softens, a nifty trick in itself. But not the one we're ultimately interested in.
What this demonstrates is that High-Pass and Gaussian Blur are simply two sides of the same operation. Gaussian Blur is the low pass filter, blocking out all the fine detail and only allowing the largest, most basic areas through. The level of detail is based on the radius setting.
The High-Pass filter works the same way, but in reverse. It blocks the lower bands of detail and only allows the higher bands of information, the fine detail, to pass through.
The radius is the key. Different kinds of detail are maximized at different radius settings. The trick will be to isolate those different "frequencies" of detail, emphasize the ones we want, and minimize the stuff we want to get rid of.
Let's return the image now to the state it was in before we applied the High Pass Filter. You can go the the history palette and click on the level just before the one that says High Pass Filter.
(If, for some reason, your history has gotten scrambled, just set up your layers as before: original on bottom, blurred layer next, and a copy of the original on top)
This time around, we're going to set our blending mode and opacity BEFORE applying the High Pass Filter.
Go to the top layer (the copy of the original) and set blending mode to Linear Light, and opacity to 50%. It should look something like this.
Open the High-Pass Filter and and pull the radius slider all the way to the left. The value is .1 pixels.
The image to the right is magnified to 400%. Note how the level of detail changes as you increase the radius. Your readings will vary somewhat, depending on the resolution of your image, but the same thing will happen as you raise the setting. Different levels of detail will become more pronounced.
What we are left with, after this somewhat complicated demonstration, is really a simple idea: Various settings of the High Pass filter emphasize different levels of detail. Some of that detail is desirable. Some, we want to get rid of. It's actually pretty simple to accomplish.
It's always a good idea to apply some conventional retouching to the image before embarking on this procedure; large blemishes, warts, brusies, lines and discolorations should be minimized as much as possible. But you don't need to be a fanatic about it. As I said at the outset, I performed no retouching on this image before taking it through these moves, though, as I mentioned, the skin was in good condition to start with. Sometimes, though, it helps.
Copy the original (or merged layers) into five new layers.
Name the first layer Blur and leave it in Normal mode at 100%.
Change the other four layers to Linear Light mode and set their opacities each to 50%. Turn them off for now and return to the Blur layer.
This is the step that wipes out everything, in order that we can selectively replace only what we want. We've used Gaussian Blur so far in our examples to accomplish this.
The advantage with using Gaussian Blur is that it is precise. Knowing the radius of your blur provides you with a guide for measuring the radius for your high-pass filters. The disadvantage is that it is sloppy and hard to control. Color bleeds from one area to the next and with skin that becomes obvious next to dark backgrounds, hair, clothing and shadows.
One way to get around this has been to create a skin mask like the one to the right. Then, using it as a selection, the skin gets copied to a new layer and the blur gets applied to that. This works better, in that dark regions adjacent to skin don't bleed their colors into skin tones. But it creates a softness at those edges anyway, simply from blurring. There are ways around it, but I prefer a different method altogether.
THE LENS BLUR
The Lens Blur filter was introduced in PS7, I believe, and its primary function is to mimic depth-of-field effcts. It uses a grayscale alpha channel to determine how much blurring to apply to different parts of the image. A properly applied gradient from white to black can create a believable effect of blurring increasing with distance. It is far cleaner than trying a similar operations with the gaussian blur. It adheres to the mask precisely, with no pixel/color bleed. And we're not using gradients. Our only concern is with black and white areasto blur or not to blur.
I've been using a Lens Blur in place of Gaussian Blur, with superior results. The downside: the radius settings don't seem to match those of the Gaussian blur, so you will need to work a bit more intuitively. This, as it turns out, is not too hard to do.
So, shiny new skin mask in place as one of your alpha channels (named, hopefully, skin mask, or something to identify it), let's see how it works.
When you open the Lens Blur Filter, there is an impressive and daunting array of parameters. I haven't found much use for any of them other than Radius, Shape and Distribution.
At the top under Depth Map, select the alpha channel with your mask (which, if named properly, will be easy).
Leave the Blur Focal Distance at 0. If your preview shows the blurring happening in reverse, check the Invert box.
I use the Hexagon shape because it seems to be stronger. It's not really important. Choose to taste.
ª At the bottom I check Gaussian for Distribution.
It's the Radius slider that is of interest. By now you should know what it is we're looking for with this step. Vary the setting until you have those smoothly blurred areas of color.
If you see that you are losing some shadow detail, or that you are getting some color drift from shadow areas into skin tones, click Cancel and return to your mask and paint them out. Notice how in the mask I drew, I excluded the nostrils.
Once your preview is to your liking, apply the filter. It takes a bit longer than a normal Gaussian Blur, and the result is a little strange. Sharp eyes and mouths in a smooth sea of washed out skin. But that will change.
Up until now, the steps have been pretty straight-forward, and the goals clear. Now it becomes slippery, in that there are an infinite number of ways to set the appropriate values to get the effect desired. It's a matter of taste and preference. I'll take you through the settings that I chose, and try to explain why (not always an easy thing) but it's not as important to match my approach as to become comfortable with what is actually going on. Keep in mind, your numbers will be different. In fact, were I to do this again, my numbers would probably be different. And any of these parameters can be enhanced, modified or expanded.
We have four layers to work with. The goal will be to apply the High Pass filter to each of them, with a distribution of radius settings that more or less evenly captures a full range from small to large detail. Once we determine the initial radius, we can then determine the settings for the remaining three layers by simply reducing the settings each time by an equal value. But what should that initial radius be?
Recall that with a Gaussian Blur, we knew the setting that would return the blur to its original state: it was the same as the Blur radius. We don't have the same precision here, but it's pretty easy to increase the radius of the High Pass filter until the image looks more or less like the original. Our starting radius will then be a value lower than this. (Remember, we're trying to minimize the large detail.
For this image, a radius of 30 pretty much replicated the original state. So I reduced the setting to 20 and that became my starting value, the results shown here. Compare it with the third image (the starting image) and you can see that we've already begun smoothing some of the blotchiness away.
Our ending value will almost always be a radius of 1 pixel (occasionally less, but not often), so it's simple arithmetic to figure out the intervening settings. I chose 20, 13, 8 and 1 as the settings for each of my High-Pass Layers.
This is how it looks after all four layers have had the High Pass Filter applied. Things look pretty ugly; but now we start playing with the layer opacities.
We will drastically reduce the opacities of the layers with the largest radius settings, and increase the overall effect of the layers with the smaller settings.
Here is the arrangement I ended up with. The layer names reflect the High Pass settings each used. I arrived at this by pure trial and error and might well have chosen different values. But these work. Trust your eye.
The opacities are as follows:
I then grouped the two top and two bottom layers into their own layer sets, and adjusted the opacities of these as well.
Low End: 20%
High End: 75%
Finally, I applied a Gaussian Blur to each of the High Pass Layers, to return things to a realist focus. The blur radius increases as the High Pass radius increases. I stayed conservative. Experiment to see what works.
And, finally, I grouped both layer sets, along with the blur layer into a master layer set and applied our original skin mask to everything.
NOTE: there's nothing hard and fast about using four detail layers. Any number could be appropriate. You might want to cluster detail layers towards the smaller bands. Each image will have different characteristics and requirements.
And the addition of a pleasant facial glow (also covered in Part III) completes the effect.
For all the convoluted explanation, the actual procedure takes about 15 minutes.
The variations are infinite. Depending on the opacities of the various layers you can create a crisply focused look or a soft fuzzy look. If the underlying image is fairly well retouched, the entire set of smoothing layers can be reduced in opacity to blend back into the original. Another approach would be to cluster the High-Pass settings towards the smaller end, rather than spacing them out evenly over the full range.
They all add up to the same thing: bluring the big chuncks out of existence, keeping the small bits and making them the focus.
Discuss this technique on Dgrin, here.