This is a technique that's been adapted from film, where it involves lots of cleverness with multiple monochromatic exposures and differently-coloured filters. Digital cameras dispense with the need for cleverness, opening up this sort of thing to any old pudding.
All you need is a scene which features both static and dynamic elements. The end result will be an image in which any moving object has produced brightly coloured frills, trails, or ghost images of itself, viz.:
It works best if you use a tripod, but holding yourself really still is also acceptable. It just makes life slightly more difficult later on. You want three different exposures of your image. The amount of elapsed time between them necessary to create a good image will depend on the subject. I just tend to hold the button down and get eight or nine shots and then sort them out later. Good subjects include traffic, busy streets full of people, waterfalls, rapidly-moving clouds, trees blowing in the wind, and especially the patches of light and shadow created by said trees. If you fill the frame with nothing but moving stuff it can get a bit trippy, though, so like I said, try and include some static elements as background.
Here are the three fairly dull shots of Woodlands Road that I combined to make the image above:
These link to the original files if you fancy having a go straight away
Colour digital images are composed of colour channels. Initially, these will be labelled red, green, and blue. The camera's sensor[1] has its light collecting parts divided into sites which record 'red' light, sites which record 'green', and sites which record 'blue', where each of these is a distinct set of wavelengths. It's all done with filters. Three separate images are recorded in monochrome, specifying how much brightness each colour contributes to the image. These are then combined to produce a colour image.
For example, the red-orange sign to the right should look very bright in the red channel, and much darker in the blue and green channels. To have a swatch, select the channels palette in Photoshop and then click on each channel in turn.
The red, green and blue channels, respectively. Note manky blue channel
Notice that bright objects are bright in all three channels, while dark objects appear dark in all three. The green trees are bright in the green channel, but—hold the bus—also in the red channel. There are good reasons for this, and they have nothing to do with red trees. If you're that bothered, there is a footnote.[2] Crucially, each channel has
all the same stuff in the same place, just at different levels of brightness. That's what we're going to change.
Select your first image. Go to the channels palette, and click the red channel. Now hit CTRL+A to select it all, and CTRL+C to copy it. Select your second image. Channels palette, click red channel. CTRL+V to paste.
Unless you were very patient and did things properly with a tripod, everything should have just gone a bit 3D-glasses. Make sure you've got the red channel selected (CTRL+1), then click the little box just to the left of the RGB channel. An eyeball should appear, meaning that you're viewing all three channels but only editing the red one.
Use the move tool (V) to click the red channel and drag it till it lines up with everything else. Things to look for are high-contrast edges in the static elements of the image. It should be relatively easy to align these. If everything else still looks skew-whiff, you may have to rotate the channel (CTRL+A to select, then Edit>Transform>Rotate). If you tilted the camera enough to mess with perspective or moved closer to/further from your subject between shots, it's probably easiest to give up and try again. It'll be quicker, trust me. Use a tripod next time.
Once everything's lined up, only the stuff that moved between shots should look different. Any parts of the new red channel that are lighter than in the original will have turned the corresponding parts of the image red. The darker parts will look cyan. So far so good.
Now select your third image (the one you haven't touched so far), and hit CTRL+3. This selects the blue channel without you even having to click on the channels palette first. Useful, non? Hit CTRL+A to select and CTRL+C to copy. Now go to the blue channel of your second image[3] (the one that now has exciting colour ghosts) and hit CTRL+V. You should get the same effect as before, only now with blue and yellow instead of red and cyan. Line things up using the move tool as before, and review your work. Where the red/cyan ghosts and the blue/yellow ghosts overlap, you'll get greens and magentas into the bargain.
That's it. If you want to make the effect more pronounced you could try popping the saturation up a little. I tend to make a duplicate layer (CTRL+J) and set it to 'soft light', but that's just me. Experiment. What works for one image may not necessarily work for another.
Examples
Some things to consider:
- What happens if you apply this technique to three images of a static object that's been lit from a different angle each time?
- What if you pan to follow a moving object and take three shots in quick succession (so the subject appears static while the background changes)?
- What if you try using a different colour mode such as CMYK or LAB rather than RGB?
- What happens if you combine channels from completely different images?
- What happens if you select one channel (green, say) and flip it about its horizontal or vertical axis?
- What happens if you make an image out of three of the same kind of channel (say, three blue channels)?
I'm going to show you that last one because it's one I've tried recently and it's usefully illustrative.
This image of Ryan Woolworths juggling is okay, but it could use a little more pizazz. The background is maybe a little too saturated for the effect to really stand out. Let's go back to the original pics it was made from. This time we're going to lift the blue channel from image 1 and paste it over the red channel of image 2. Then we're going to lift the blue channel from image 3 and paste it over the green channel from image 2.
Everything that was motionless has become monochrome. The reason for this is key to the whole thing:
pixels which have the same brightness value in all three colour channels will be monochromatic. By mixing equal amounts of red, green and blue light, the result is white. By reducing the brightness equally in each channel, we get a progressively darker shade of grey. Zero brightness in all channels results in black. A colourful object will result in more light arriving in some photosites than in others, leading to differing brightness values in different colour channels, hence colour. The blue channel for each of these three images was near enough identical for most of the background, so only the areas where they differ (ie, where there was motion between shots) will appear colourful.
Maybe
too colourful. It's very distracting on the face. Juggling should be about arms and balls. I'm going to pick the blue channel with the most intense face, copy it, then switch to the layers palette and paste it as a new layer. Then with the application of a layer mask I can paint out the colour from the bits I want to be straightforward black-and-white.
Blatantly a different image I made much earlier
Hurrah. Have fun.
[1]Provided we're talking about Bayer sensors. If your camera has a Foveon sensor then you probably know more about this stuff than I do. If you don't know what type of sensor your camera has, then unless you've been bought a Sigma DP1 by a well-meaning relative then you've almost certainly got the Bayer-Mosaic-Filter sort. [2]There's some yellow in the leaves, especially in the just-off-centre tree in the background. Yellow, in RGB images, comes from mixing red and green, so while the tree is predominantly green (and hence brightest in the green channel) the yellowish hue means that some of its reflected light is collected by the red part of the sensor. Yellow objects look black in the blue channel, just as magenta objects in the green channel and cyan objects in the red channel. Science-wise, these colours are opposites. 
Opposite colours, yes? [3]CTRL+3 will work here, too, but since you clicked that eyeball earlier it won't look like anything's changed. 
