Stereo Pair Photography (the low cost way)

Written by Paul Bourke
July 1999


Introduction

Stereo photography involves taking a photograph from two positions, these correspond to two "eye" positions. The two cameras cannot just be arbitrarily separated and point in roughly the right direction. The separation depends on the distance of the closest objects in the scene and the degree of stereo one wishes to achieve. The cameras should be angled inwards * so that imaginary rays projected into the scene intersect at a depth that is intended to be at the projection plane. Objects that are closer than this intersection point will appear to be closer to the viewer, points behind this intersection point will appear to be further from the viewer. There is a limit to how close objects can be made to appear and still provide comfortable viewing, a general rule of thumb is for the camera separation to be around 1/20'th of the distance where the camera rays intersect.

The two photographs are presented independently to each eye by any number of techniques. Some people are able to focus their eyes on a more distant position and view the two images (stereo pairs) unassisted. Most people prefer some assistance in the form of a stereoscope or more recently by stereo computer graphics devices. Whatever the means, the result is the appearance of depth....the image appears to be 3 dimensional. The technique was very popular in the late 1800's and early 1900's where it was embraced by the relatively new photographic technology. The following stereo pair was taken of the USS Oregon with a specially design stereo camera in 1898. These images were originally designed to be viewed with stereoscopes which were developed in the early 1800's. The popular Wheatstone and Brewster stereoscopes used prisms and mirrors to present the left and right images independently to each eye.


Left view

Right view
Example 1

The following example was captured using a high definition digital camera. The clock tower is part of the town hall on the corner of Glenferrie and Burwood roads in Melbourne, Australia. One of the problems of using a single camera and taking two shots is that any movement in the time between the shots results in viewing degradation. For example in the following images, the area around the flag seems confused because it had moved between the two images. The two camera approach needs to be abandoned for the stereo photograph of all but still scenes, even moving clouds can ruin the effect.


Left view

Right view

The question of the separation of the camera is not as straightforward as one might imagine. It depends on the size of the image in relation to you distance from it, as well as the degree of the 3D effect one might try to replicate. In practice it is good form to take a number of shots at various separations and combine the best pair by experimentation afterwards.

Example 2

The following image is one taken from a movie recorded using two digital movie cameras. This is much harder than taking single shots, both in terms of filming and the subsequent processing. The main ingredient is a strong tripod and mounting bar for the cameras. The mounting needs to allow the cameras to slide apart from normal eye separation to perhaps 1 meter separation. It needs to be able to be precisely leveled. The hardest part is turning the cameras so they focus at the depth of interest, normally the cameras are mounted on dials with a fine gear arrangement that turns the cameras together (but in opposite angle directions, one turns clockwise while the other turns anticlockwise). As with a single camera it is necessary to have a cross hair arrangement in order to point the cameras at the same point.


Left view

Right view

Given that the cameras are aligned properly during recording, if the cameras are not synchronised as is the case with two independent cameras, then the two captured video streams need to be aligned in time. For the exercise discussed here this alignment was done in the stereo viewing software itself but it could just as easily be done using one of the many commercial digital video editing packages available. It is assumed that the clocks in the two cameras are precise enough not to slip in time. The alternative is to use time locked (genlocked) digital cameras, this is a much more expensive option than using independent "consumer grade" cameras.

Viewing

There are many ways of viewing stereo pairs. A popular method in the early days of computer graphics was to use red/blue glasses and superimpose the two views into one image after turning the left and right views into the appropriate colour. So for example if the glasses had blue lens on the left and red lens on the right, then the left eye image would be shaded blue and the right eye image shaded red. This images resulting from this general technique are normally called anaglyphs. The example below is from the mars lander which took a large number of stereo pairs from the surface of mars.


Left view

Right view

Anaglyph

The obvious problem with this method is that only grey scale images can be viewed. Another approach is to project the image with say two different polarisations of light and then the user wears glasses with the appropriate filters. Horizontal and vertically polarised light is frequently used but it doesn't allow head rotation. A better technique is to project with right and left circular polarised light. The projection simply involves two slide or movie projectors with filters over their lens, it is important though to project onto a screen that doesn't destroy the polarisation, metal projection screen satisfy this.

More sophisticated techniques involve high refresh monitors that alternatively draw the left and right eye images. The viewer then wears glasses of some kind that allow the eye to see the images it is supposed to see. One approach is to use LCD shutters that alternatively turn from transparent to opaque, the shutter for each eye is synchronised to the images being displayed on the monitor. This synchronisation was first done by a cable but is now usually accomplished with an infared emitter allowing greater movement and comfort by the wearer. With the advent of reasonably low cost shutter glasses this technique is becoming popular in cames for the home computer market. To be really successful and to minimise artifacts it is necessary to acquire a monitor with fast phosphor and LCD shutter with a fast response. Without these there can be significant ghosting, some of the left eye's image for example is seen by the right eye. An example of a viewing arrangement is shown below, the red light on top of the monitor is an infared emitter that synchronises the glasses with the alternating images on the monitor.

Source

C source (glimage.c, glimage.h) or (glslides.c, glslides.h) for displaying stereo image using OpenGL and the glut library. This code is definitely not general but was written to form a "proof of concept". It displays two "raw" images in the left and right frame buffers, the image must be 800 by 600 and the monitor is assumed to be working in stereo at the same resolution. Other configurations should easily be able to be supported making small changes to this code.

Toe-in and offaxis stereo

The method of turning in the two cameras is known as "toe-in". It isn't the ideal method to use since it introduces various distortions, most noticeably vertical parallax increasing out from the center of the image. It was discussed here because it doesn't require modifications to a standard camera or video camera.

The better way to photograph stereo pairs is with what is known as offaxis projections, unfortunately this requires either special lens/film arrangements or in the case of CCD cameras the CCD needs to be shifted horizontally on both cameras (in different directions). When using an offaxis camera the cameras are mounted parallel.

Another option is to photograph a slightly larger view than is intended with parallel cameras. The offaxis projections are created by aligning the two images and clipping off the unused bits, this will involve removing a strip on the left of the left pair and a strip on the right of the right hand pair. The exact amount to cut off can be determined by eye by visually inspecting the alignment or calculated by straightforward geometric considerations. Photoshop is particularly good for visually aligning the pairs because they can be superimposed with one slightly transparent, no doubt other image manipulation packages can achieve the same result.

References

Ferwerda, J.G.
The world of 3D, a practical guide to stereo photography
Netherlands Society for Stereo Photography
1982

Lipton, L.
Foundations of the stereoscopic cinema
Van Nostrand Reinhold Company
1982

Marshall, G. and Gandland, G.
Method of making a three dimensional photograph
United States Patent Number 4,481,050
November 6, 1984

Okoshi, T.
Three dimensional imaging techniques
Academic Press, New York, 1976

Waack, F.G.
Stereo Photography
The Stereoscopic Society, London, 1985




Manual Creation of Stereo Pairs

Written by Paul Bourke
April 2001


While most stereo pairs are created with either stereo cameras or by computer, it is also possible to manually edit a normal photograph and introduce stereographic 3D effects. The basic requirement is the ability to selectively (generally using computer based image editing software) choose objects that should appear in the foreground or background. These objects are displaced horizontally on each stereo pair image, the amount and direction of the displacement dictates the depth and whether it is in front of the image plane or behind the image plane.

The technique used here will be discussed with an example, the original image is shown on the top-right hand side. The idea isn't difficult but can be tedious and for more complicated images can require a sound methodology and careful planning.

The first processing in this example was to remove the noisy appearance between the galaxies and stars. The plan is to keep the two galaxies in the plane of the image (0 parallax), the white stars will be brought towards the viewer (negative parallax) and the smaller reddish stars and galaxies will be moved into the distance (positive parallax). If the noise in the plane of the galaxies isn't removed it will form a strong appearance of a plane.

The different objects are marked (on a separate painting layer) with different colours depending on their intended depth. This method was chosen simply because it is easy to then select the regions by colour using the tools in the particular image editing software being used (PhotoShop). The red objects will be brought closest to the viewer, green further away but still in front of the image plane, blue and yellow are progressively further away. The more depth variation used the better the end result but it greatly increases the time consuming nature of the process.

At this stage two images are created, one for each eye. The appropriate pieces are copied/pasted onto different layers within these two images. The layers are then shifted left and right as appropriate. For example the red objects are shifted to the right in the left eye image and to the left in the right eye image (negative parallax). The yellow objects are shifted left in the left image and right in the right image.

The amount of separation requires some experience. Objects that are to be at infinity are separated by the intended eye separation, objects brought towards the user shouldn't be separated by more than this eye separation. The amount of separation is also dependent on the way the stereo pairs are presented to the viewer, some systems can sustain more separation than others. It is possible to leave one eye image as it is and just shift the layers in the other image by twice the amount, the only issue with this is it increases the chance of problems at the left and right edge where there isn't enough information or where an object on one stereo pair disappears off the the side during the shift.

  



Final left eye image
   Final right eye image