So, my wife having left me for some high-speed training this week, I want to make use of the free time by embarking on a short series to try to make up for where my previous attempts to tackle the subject have fallen flat.
Before we get started, though, this is a very complex subject, and as such, I am very much going to be glossing over a lot of the details in the first few posts in order to convey the basic principles. The truth, as they say is ugly, and so we'll save it for when you have the foundation to tackle it properly.
For most people, our familiarity with frequencies is in terms of sound. We all know that birds chirping, glass breaking, and children screaming are (typically) high-pitched (or high-frequency) sounds. Equally, we associate sounds like explosions, fog horns, and books dropping as being low-frequency sounds. But what does that have to do with images?
Well, we have to go back to high school physics for that one, so bear with me here - it's been a while for both of us. Do you remember what a low frequency sound 'looks' like, with a longish wavelength, or period? The figure at right (generated with ASU's J-DSP Editor) shows a graph of such a generic low frequency tone.
And what good would an arbitrary 'low' frequency tone be without a complimentary 'high' frequency tone? The figure at left shows just such, having a frequency which is twice that of our first sample. [Don't worry - I know this is boring, but it is leading somewhere good!]
Now, no one likes to sit around listening to single-frequency hums day in and day out, so what is happening physically when we mix two sounds together? You might remember that the two components can combine either constructively, emphasizing one another, or they can combine destructively, canceling one another out. The figure at right demonstrates what happens when we combine our low and high frequency sounds from above. Note how they combine constructively and destructively, depending on their relative values.
Let's trying applying this in a more visual sense. To do so in Photoshop, I constructed the figures at left, which consist of nothing but vertical bars evenly spaced across the screen. In a second layer (also shown at left) within the same document, I created another series of bars, this time twice as wide as those in the first. These should be considered as analogous to the equivalent sound waves which we looked at above. If we combine them in Photoshop (a process which we will go over later), we should get something similar to the combination of sounds from before. In fact, the third figure reflects just that - an eerily close replication of the sound pattern. Pay close attention to the way that the highs and lows combine just as they do in the audio signal. And while this example is simply one-dimensional and wholly contrived, it is a process which occurs across as many dimensions as we feed it - highs continue to build on other highs, and to cancel with lows.
At this point, it would be reasonable if you're thinking, "Gee, it's great that you can compose these frequencies like that Sean, but I'm not making photographs - I'm retouching them." Here's the magic part. Just as every sound which your computer microphone records can be broken down into its component frequencies using processes which are the reverse of what we did above (we'll talk about them later); we can use Photoshop to break images down into their component frequencies. Let me show you what I mean.
The triptych below shows a breakdown of a shot of DC United player Chris Pontius as he gains possession during an MLS match (yes, it's a 'real' photo). On the right is the image as output from Lightroom. One the left, I've used the Gaussian Blur filter to show only the lower frequency portions of the image; in the middle, only the high frequency portions. By combining the two back together, we can recreate the original. I will tell you right now that we can do this very, very accurately in Photoshop - at least as accurately as we can switch color modes.
This next image does it again, but this time I've broken the image into three different segments, combining back to the same source image. For now you'll have to take my word that we can do this as many times as we like for an unlimited number of separations, and really limitless possibilities for retouching.
Over the next few blog posts, I'll get into the hows, the whys, and above all things the details of this, but for a moment let sink in what we just demonstrated. Where many of us grew up in the retouching world with Margulis and Krause teaching us the 'revolutionary' idea that an image could have 10, 13, maybe 20 or more channel-based representations of itself; this idea represents the ability to create as many more permutations as we could ever want. You can look at an image not just in terms of additive or subtractive color; not just luminance and chrominance; nor even hue, saturation, and lightness - no, you can combine these with size; even with shape. We have a lot more to talk about, but all in good time. If you're too anxious to wait, head on over to ModelMayhem to read the "HighPass Sucks (+ solution)" thread which got a lot of this hubbub started; otherwise, I'll hope to see you again here soon.
Addendum: Much of the above was written hurriedly. If there are typos, I would appreciate your help in identifying them. Part of the rush has been that it was originally my intention to make this a video tutorial series, but I'm embarrassed to admit that I no longer know of any (free) utilities for generating and mixing constant audio tones (demonstration of mixed audio being the biggest boon in moving to a video format). If you know of such a utility (which is also GUI'd and easy to use), please drop me a note!