Viewing Stereo Images
These
images are
combined using the anaglyph
method (using coloured glasses), since most people have a hard time
getting cross-eyed
viewing to work. Also, everyone has a
pair of paper glasses they got from a cereal box or the latest SI
Swimsuit issue. If you don't have any glasses, you can get
some cheap paper ones here or some really good plastic ones.
What you need is red-cyan glasses, with red on the left eye. To get the most out of these images, view them in a dim room and step back and forth until the effect seems at its best. Viewing them against black also helps. Sometimes I like to zoom in/out the picture and pan around, you can try ACDSee to view them as it's really good with zooming in/out on black.
Making 3D Images - A Little Theory
The
principle of Stereoscopy is to take two pictures that are taken with a
slight horizontal offset. The resulting images are each displayed to a
single eye. The brain does the job of merging the two together in a
single image. A good 3D image will trick the viewer into perceiving
depth in an otherwise flat image.The offset between human eyes is on average 60mm, this distance is a good starting point when shooting images that you intend to combine. If you want to shoot subjects that are farther away, you need to increase the offset to preserve the depth perception. Same with objects that are really close; you need to decrease the offset, or else the brain won't be able to merge the images together. Typically the rule of thumb is that the offset should be 1/30 of the camera-to-subject distance (e.g. a subject that is three meters away should have an offset of ten centimeters). This rule is not set in stone. You can increase the distance to augment the "3D effect", but you run the risk of making the images too different for the brain to merge them.
In case you're really interested going deeper into the subject, here is a very technical manual that covers all the theory behind stereoscopy (elaborate math included).
Single Camera Stereoscopy - An Early Attempt
My friend Francois and I wanted to experiment with stereoscopy with digital cameras for a while. Since we worked in 3D and VFX, we did get a chance to do some stereo with computer generated elements. I did try experimenting with a single camera on a sliding plate. In this case I slid the camera left, took a picture, slid it to the right, and took a picture. I composited the result in Photoshop and was pleased with the result. I even tried adding some composited text and it looked pretty good.
The only issue is that I wanted to take pictures of people, and I couldn’t since they moved during the time it took me to slide the camera. Luckily, my friend François happened to own a camera and lenses that matched mine (350D / rebel XT with 50mm f/1.8 and Canon 10-22mm). This meant that we could finally try dual-camera stereo imaging!
Dual-Camera Stereoscopy
The Rig
The challenge was to get the two
camera lenses as close to one another as possible. Typically the
distance between human eyes is 60mm, so we usually want to match this
distance between lenses to get a realistic effect at a reasonable
distance. Realistic is fine, but sometimes with stereoscopy, you want
to enhance the effect. In this case we managed to get 100mm of
separation, which gives a great effect but isn’t too
exaggerated. If you try to increase the distance between cameras too
much, the brain just gives up trying to reconnect the images.
I created a custom
rig to line up the two cameras. I used an aluminum bar commonly found
in hardware stores and cut it into three equal pieces. It can be cut
easily with a hacksaw. I stacked the three pieces, clamped them and
drilled three holes on the drill press (the aluminum is soft enough to
use regular wood/metal bits). I filed all edges. Gaffer tape was used
to prevent the strips from slipping. 
Each
camera is attached to its own strip using a screw that has the same
size and thread as the camera’s tripod mount. These are flat-head
¼ inch diameter 20 thread screws
(typically the most common ¼ inch screws available in
hardware stores). Since I had a hard time finding the right length (3/8
inch) I got longer ones and cut them. If you decide to do this,
remember to screw on a nut before cutting the bolt. This will
straighten the thread when you unscrew it and will give you something
to clamp onto when cutting it. (note that you can use a hacksaw or a
grinder). Remember to file the cut edge and bevel it a little bit. 
The
middle hole on each bar should be countersunk with a larger diameter
bit so the bolt heads will be a little recessed. Since I
couldn’t drill the middle holes enough to completely recess
the heads, I used a third bar between the two others as a spacer. I
enlarged the middle hole to allow clearance for the bolt heads.I then screwed the bars to each camera using the shortened bolts. All bars were then attached together with longer bolts and wing nuts. The result was quite rigid.
The Remote
The
details of making a remote can be found on the connections
page. In this case I ran wires from the remote plugs on each camera and
connected them to a DPST switch (which is a double pole switch
–two switches in one). The switch connects the tip and the
base of the 2.5mm jack – this triggers the shutters. I then
tucked the switch and the wires in an empty enclosure. The latter version included an AF button to make sure both cameras were focus locked on the subject before the shutter was triggered (thus preventing any focus delay).
Shooting Technique
Make
sure both cameras have exactly the
same settings, all set to manual (the eye is very sensitive to
discrepancies between images. This includes WB, saturation, sharpening,
everything. We used RAW to avoid any problems, and this affords us a
lot more latitude when post-processing the images. If you plan on using
flash, use a slow shutter speed to make sure that both shutters stay
open until the flash has fired. Also, make sure the date and time
matches exactly on both cameras. This will make it a lot easier to join
the pairs, as you just need to sort the folder of images by date.1st Attempt – Available Light
This
is a test of the original rig,
using available light. We quickly realized that using a 50mm was much
better than 10 or 20mm, because at that focal length, you need to get
really close to the subject and the difference between the points of
view is too great for your brain to make the connection. Here is an
example at 10mm. What came of this exercise is that since we only had control over the shutter and not AF, the focusing time between cameras varied and resulted in shots with up to a second of difference (way too much). We could have used manual focus, but it’s really difficult on two separate cameras to get the same focus, and the 50mm 1.8 has a notoriously flimsy focusing ring. We set forth to include focus control on or remote so we could lock the focus before we triggered the shutters.
2nd Attempt – Indoors With Flash
This
time was much more
fruitful. I decided to use some strobes, but the flash sync was an
issue. I got the idea to connect the flash sync to each camera in
series. I hade a basic hotshoe adapter. It's quite rudimentary,
but it works. It's made out of two pieces of plastic, nails and a few
wires. for those who don't understand the principle behind a hotshoe, check this out.
In this case we also used a tripod, which definitely helped getting more accurate framing of the subject. We also shot on black, which helps convey the sense of depth. Another thing that helps is when the subject doesn't touch either side of the frame (this disrupts the 3D effect near the edges).
All in all, we got some pretty good results. We tried different shots, which can all be seen in my gallery.
Combining
the Pictures
In
this case Photoshop was used, instead of some of the specialized
software you can find on the web. This provides a lot more latitude,
and you can even retouch the final composite. I was looking for a
simpler solution than replacing the red channel (as is mentioned in
most other tutorials). Instead I ended up creating two layer sets with
layer effects to divide the channels. This allows for better color
correction when the shots are combined, and allows for easy conversion
from color anaglyph to Black-and-white anaglyph (using the channel
mixer - I'm a stickler for good b+w tone). I'd like to share my Photoshop template. Replace the left and right layers in this template with your own images. Remember to set the convergence on the object you want to be in the plane of the monitor (i.e. line up what you what to appear at the depth of the monitor surface). Shift-dragging the images on the document will center them in the frame. You can then play with the adjustment layers to get the contrast just right. Note that if you hide the B+W adjustment layers, you will get a color anaglyph.
If you plan on trying this, let me know how it turns out.