In this chapter we outline the theory of registration and the stereo window.  This is followed by a discussion of some tools to help get it done efficiently and effectively.  The emphasis here is on making prints for viewing, or making stereo pairs for the web.  Registration methods for projection are discussed in chapter 7, though the basic theory is discussed below.

Before presenting a stereo pair it is necessary to align the images carefully.  Film slops around inside film cameras (unless you use a very specialized scientific-grade pin-registered camera).  Images from twin digital cameras may be skewed by camera mounting errors or by individual cameras that have their CCD's in slightly different positions, for example.   Also, pointing errors on a slide-bar come into play if the view directions are not exactly parallel, or if there is some rotation of the one of the cameras about the view angle.  Alignment errors are exacerbated with telephoto lenses.   Other sources of registration errors are:  printing done at a photo finisher,  the digitization of film for home printing, the mounting of slides in plastic frames by your processing lab.  All these procedures can lead to small twists or translations of the images with respect to the edge of the print or the frame of a slide mount.

This misalignment of the images can cause severe eye strain and make it difficult to fuse the image pair, no matter what the presentation technique.  For slide or digital projection, where small misalignments are magnified up to large displacements on the screen, the situation is particularly severe.  An image pair should be carefully registered before printing, and slides may need to be re-mounted with careful attention to each image's position with respect to the frame.

REGISTRATION THEORY

Here is a simplified discussion of registration.  For a more details see Ferwerda, J. "The World of 3-D:  A practical guide to stereo photography."

Fig. 5.1  Image Registration There are three ways to move, or register,  the right image (blue) with respect to the left one (red): Horizontal Translation in x. Vertical Shift in y. Twist , or Rotation, in q. After any manipulation the area remaining in common will be smaller.  Hopefully you don't have to make large movements and the area lost will be small.  In the example on the left, the right image was elevated,  translated to the right, and rotated CCW.  The green area shows the left over common area that can be used in the stereo image.

The steps involve in getting the images in proper alignment are:

1) Remove any vertical error.  This is done by moving the right image in y until common points (a tree top, a mountain peak on the horizon) are at the same height.

Fig. 5.2a  A misaligned pair.  Red is L and blue is R.  Both the near object (the flower) and the distant grass are high for L (red).  This must be corrected to enable comfortable and effective viewing.  Your eyes are separated horizontally and your brain can accommodate and fuse a horizontal disparity.  Your eyes are not separated vertically, so anomalous vertical separation in the image can cause discomfort or brain pain.

Fig. 5.2b  After a y-shift of the blue (R) image, it comes into line vertically with the red one.  The peak of the petals are at equal height.  So are the blades of grass in the background.  For the near-horizontal blades, red and blue overlay each other and these blades turn gray.  Note that horizontal shifts remain in varying degrees in the near and far field.  Theses shifts are the stereo effect!

In the above example note that a straight vertical shift of R led to vertical alignment all across the picture.  Perhaps the film was riding high in the R camera.   The fact that we were able to align the images all across from left to right  means that there was no twist error in the originals nor was there a large tilt error on the twin-cam bar.

2)  Correct any rotational error.

If an alignment of the peaks of the petals in figure 5.2b had left a misalignment of the grass in the background on the left, this would need to be corrected by rotating the blue image as well as vertically shifting it.  Up-Down camera tilt errors (one with respect to the other) cannot be exactly eliminated because they affect the background more than the foreground, and the errors at both locations cannot be simultaneously removed with a simple vertical shift of the planar images.  However, most errors, if not too large, can be substantially abrogated.  Vertical shift errors of film in the camera or of the camera on the bar, or twists about the line of sight, can be exactly removed (at the expense, of course, of losing some of the image area, as shown above).

In summary, move the R image to get most of the same features in the image at the same height.  Do this with isolated identifiable features that are at a common physical point in the two images.  In fig. 5.2 we used the peak of the petals, or where blades of grass cross in the background.  In this example we moved R with respect to L, but you could do it the other way around.

3)  Correct any magnification error.

Sometimes, especially when using two cameras, two zoom lenses, or a staggered mount where one camera is slightly behind the other, one image will be slightly magnified with respect to the other.  The magnified image should be resized downwards until equal objects have equal height and width.  This is very difficult to do with film.  Film can be translated or rotated, but resizing means re-shooting dupes.  Digital processing of digital images is superb here.

4)  Adjust the horizontal separation (the "stereo window") by translating R with respect to L to get the desired 3D effect.

THE STEREO WINDOW

When you view a 3D image, the common edges of the prints, or the common frames of the slides (as seen in a viewer, or when projected), gives you a reference distance.  For example, when the frames of slides are lined up to be coincident on a projection screen, the distance to the screen itself gives you reference location, and the overlapping frames define an opening that contains the image.  It is so many feet away.  Same thing happens with a L & R pair of prints.  The uniformly colored "paper" (black on this website) making up the canvas surrounding the prints, also defines a location.  The rest of the image is then perceived relative to this reference opening that is called "THE STEREO WINDOW".

RULE OF THUMB:  Since the edges of the frame overlap coincidently, parts of the stereo pair that also overlap coincidently will appear to be at the same distance from the viewer as the stereo window.

Let's see if we can make sense out of this.

Fig. 5.3  SCENE OF A FIR TREE For simplicity, consider a simple scene with a big fir tree in the front.  As argued in chapter 1, the LR pair of images will contain the fir tree offset to the R in the L image, and offset to the L in the right image.  Let us assume all the necessary corrections have been made.  The top of the tree is at the same elevation with respect to the bottom of the image, and the trunks are aligned so that they are parallel (i.e. any rotation error has been corrected). As discussed in chapter 1, the big fir tree in the foreground appears shifted to the right in the left (L) image.  Similarly, it appears somewhat to the left in the right (R) image.

Fig. 5.4  In Front of the Window. Suppose we just overlay the two pairs from figure 5.3.  You get something like this.  As you look at this pair (in a slide or print viewer, or with prism glasses), you see the frames lined up.  Well, you should see the frames lined up.  That is, we assume the top and bottom borders of the prints are parallel, or the slide projectors are set up so the slide frames overlap, or the slides are put in a slide viewer so the frames are centered on each eye, so that the two images at the bottom of figure 5.3 and overlay them. As you look straight ahead, the right eye perceives the tree to be off at an angle to the left.  The left eye perceives it to be off to the right.  Your brain merges these two perceptions and places the tree at the common intersection of the two rays - which here is IN FRONT OF THE WINDOW. Sometimes this situation, where the rays (the red sight-lines in the figure) cross before reaching the film or print plane, is called "negative parallax:.

Fig. 5.5  Behind the Window. In analogy with Fig. 5.4, now suppose we translate the L image left, until the tree goes past the R - tree.  After doing so, we crop to a common frame and put the images in the viewer.  Now as the person looks at the picture, the rays do not cross in front of the plane of the image, rather they intersect in back of it.  This is where the brain will perceive the fir tree.  The rays intersect BEHIND THE WINDOW.

Fig. 5.6   At the Window. The obvious last case is where one translates the L image until the peaks of the tree (a common physical point - both images have a tree crest) line up.  Now the rays obviously cross at the location of the stereo window. Thus.  To put an object flush with the stereo window, translate the images until that common physical point coincides. THIS IS THE USUAL WAY OF ALIGNING STEREO PAIRS.  THE NEAREST OBJECT IS PLACED AT OR SLIGHTLY BEHIND THE STEREO WINDOW SO THAT YOU LOOK THROUGH THE WINDOW OUT ONTO THE SCENE. OF COURSE, EXCEPTIONS ARE ART (too).

In summary, 

TO SET THE STEREO WINDOW:

a)  Translate the images horizontally relative to each other to put a feature flush with the stereo window.  The translation should make this feature coincident in the overlay.  The selected feature should be a physical point appearing in both the L and R images.

b)  It is customary to put the nearest object in the scene flush with, or a little behind, the stereo window, so that you look out through the window into the scene.

c)  Sometimes, for effect, it is interesting to have part of the image out in front of the window.  For example a bird flies into the audience, or a climbing rope is tossed into the audience......  See figure 5.4.   Do not overdo this.  Save it for special cases.  DO NOT allow any negative parallax part of the image (that would be forward of the window) to lie up against the window itself, as this leads to confusion.   This is sometimes called a STEREO WINDOW VIOLATION.

4)   After the images are registered, the pair should be cropped to retain only those parts remaining in common (see Fig. 5.1).

a)  Input the left and right images as digital files.  If you use digital cameras this is easy.  If you use negative or slide film, then you first have to digitize the film.  Many scanners are available to do this.  The excellent LS2000 is cheap now on e-bay as it has been superseded.  Many other scanners in the $400 range are also good.  Look for 10 bits per color, at least, and Dmax of 3.5 or more (the higher the better).  CLEAN FILM CAREFULLY BEFORE DIGITIZING.  Dust specks and scratches are more annoying in 3D than in 2D.  Scratch and Dust removal software helps but is not a cure-all !!  Applied globally, such algorithms will often soften the best parts of your pictures.  

b)  Assuming you have input the left image and the right image, a good registration program should allow you to view a red-blue overlay rendering of the pair, and have a facility to shift one or the other (or both) as in figure 5.1.   After you have done all the shifts, a preview feature (giving a parallel viewing LR full color rendering, or a shutterglass script) is helpful to check that the alignment is giving you what you want.

c)  Crop and resize the images and build whatever output files you want.  Outputs are typically comprised of individual L and R's for printing individual images or making individual slides, composite parallel or cross-eye layouts, or shutterglass .jps files.

The Stereo Image Factory  interface looks like:

Fig. 5.8a Registration Program Layout.   Stereo Image Factory   Left Lily, Right Lily, and Red-Blue Overlay showing the relative orientation with respect to the frames of the original images. Obviously there is some vertical error to be corrected, and then some horizontal translation to set the stereo window.

Fig. 5.8b  Preview Mode The working images are too big to view wide-eyed, so after working full-screen on the Red-Blue alignment problem (where arrows are used to incrementally move the L image relative to the right), you can look at a small parallel view pair (top) with prism glasses. The parallel view pair includes only the areas in common.  At the bottom you can see that the substantial shifting has caused a fair amount of the border regions to be lost.  In this picture it doesn't really matter because the lily is the main deal. The program allows you to crop down to the common parts and resize the image.  You can crop further to get the desired aspect ratio.

The Stereo Photo Maker interface looks like:

You can also do the registration in Photoshop but the FREEWARE program STEREO PHOTO MAKER is much easier to use for stereo registration.  Here are the steps for Photoshop:

1)  Open the L and R images

2)  Copy the L image and Paste it into R.

3)  Right double click on the Layer holding the left image.  Change blend opacity to 50%.  OK

4)  Use the position tool to slide the L layer over the R (background layer) until you get the registration you want.

5)  Crop the image to the final aspect ratio you want, leaving out the non-common parts.

6)  Increase the canvas side to about 2.2 times wide, placing the image centered vertically and on the right.  Use either black or white background.

7)  Use the pointer tool to slide the L layer (that should still be selected) over into the left canvas space.  Slide it horizontally, do not shift it vertically.

8)  Right click on the L layer again and change its opacity back to 100%.  OK

9)  Do a final resize for printing or the web, and save the composite L-R pair.  To save as a jpeg you first have to flatten the picture.

The steps for Paintshop Pro (a less costly imaging program) are identical, except that you "Paste As A New Layer" in 2), and then go to the "Layer - Properties" menu to change the opacity in steps 3) and 8).

The advantage of the Stereo Image Factory program are in its facility for previewing the image as a Red-Blue anaglyph or as a L-R pair.  In SIF it is easier to do a rotation correction, though in practice I have found that this is rarely needed.  Photoshop and PaintshopPro have better image adjustment procedures (like color correction, sharpening, dust filters, etc.).

Once the image is saved as a jpeg (or other format), you can print it on your inkjet printer.  Use the highest quality paper and resolution settings on the printer because the individual frames are small.  It is recommended that you send 500 dpi files to the printer, since prints may be viewed under magnification you want the best output you can get.  You can print directly from the SIF program, but other programs like Photoshop or PaintshopPro give you more flexibility in layout.