Make your own free website on Tripod.com

Principles of stereoscopic photography
using an ordinary camera

John Wattie

Two pictures of the same object are needed, taken from two points, just as our eyes are horizontally separated.

TOPICS

  1. Camera methods for stereoscopic pictures

  2. The easiest, Cha Cha way for anybody

  3. Stereo base, Stereo Parallax: how far to separate the cameras.

  4. Camera convergence: Complications include:
          keystone effect, crossing backgrounds, segmental stereo failure...

  5. Masking for window effect

  6. Eye poking for edge stereo failure (also links to the "falling frame")

  7. Macro-stereoscopy: a summary of close up techniques, (not all are good!)

  8. Bad technique: by not following the rules on this page, a poor stereo picture is made.

  9. Better technique: how the Pohutukawa macro stereo was made.

  10. Slide and print mounting

  11. Mapping 3D space photographically. More philosophy than photography.

     Contents using frames page


Beginners

Until you are an expert, the following may seem very obscure. No worries -  you don't have to know much for a heap of fun! The contents frame to the left allows you to jump over the boring bits. On purpose, easy stuff is mixed up with erudite stuff here, which may just mean nobody is happy, but at least shared misery is fair to all?

Actually you can take 3d pictures without knowing much at all. 
Beginners go direct to here.

Real experts, especially the mathematical ones, will find this web site too easy, or wrong. Protests can be directed to the
[guest book for nzphoto.tripod.com/].


The only differences between the two pictures of  the stereo pair should be:
  1. the shift is horizontal (called X parallax in remote imaging circles).

  2. no changes occur in the vertical direction. (Y parallax is constant in the two pictures).

  3. the stereo angular parallax between nearest and furthest object is around 2 degrees when seen in the stereoscope. 

  4. vertical edges of the frames are masked correctly for stereo window.

Click here for example

 Any other difference spoils or even ruins the stereoscopy. 

Factors commonly seen spoiling the 3d include:

  1. Tilting of one frame relative to the other, one component of Y parallax error.
  2. Different magnification in part or all of the pictures (which changes things in the vertical direction, when only horizontal changes are acceptable.)
  3. Objects at infinity have too much parallax: more than the inter-ocular distance, and so cannot be fused.
  4. Incorrect virtual stereo window. "Floating frames" or objects in front of the window frame.
  5. Excessive or insufficient stereo shift for the viewing method used. Related to this are distortions and deformations of the image. For example a circle becomes mishapen and its centre seems displaced away from the viewer. "Squeeze" "stretch" and "bulge" are terms for various deformations. Deformation from a widee angle lens used too close to the object is a parallax error and not always a stereo error
  6. Bottom or top edges not at the same level in the two pictures. 
    1. which means one picture is higher than the other, or 
    2. one is tilted relative to the other, or
    3. one is bigger than the other, or
    4. one camera was aimed higher than the other.
      (The last is  partially fixed in Photoshop by cropping and the size problem is only partially correctable by image transform, but tilt is readily fixed).
  7. Retinal rivalry occurs when an object in one window is different from an object in the other. For example, a moving cloud. The brain cannot decide what is correct and often flips between the two versions, which is disturbing.
  8. Out of focus parts are objectionable in stereo but can still be fused and may become artistic if you are inclined that way, so this is a minor "fault".
  9. Noise (grain, electronic noise, JPEG heavy compression or dust and scratches)  cannot be fused and is not good. It is suggested that a small amount of blur should be applied to noisy images to overcome this problem. A slightly blurred picture also has the advantage of better JPEG compression. Often the noise is only in one colour channel (frequently blue) and so you can blur just that channel.
  10. Differences in colour and density between the pairs are not ideal but are not disastrous. Density or contrast differences can be misinterpreted as lustre or may cause retinal rivalry.
  11. Edge and segmental stereo failure are discussed in detail later...

People spend a lot of time looking at a good stereo pair, because there is so much more information than on a single flat picture. This prolonged study means poor photographic technique becomes disturbing. A "simple snapshot" that is just glanced at in a home album is unsatisfactory in stereo.

Camera methods for 3D

A stereo camera is the obvious method - too obvious for this web site! 
The best stereo camera is probably the RBT: { information here }. The RBT is two high quality 35mm cameras joined together. With the RBT macro attachment it will also take close-up stereo. There is no way to vary the separation, apart from the macro system. RBT is unfortunately too expensive for the author!

An SLR camera fitted with a { "Lens in a Cap" } split image system. This is great for family snaps but a bit limited for serious stereo. Unfortunately the stereo window is wrong: infinity instead of 2 meters away. The stereo base is small, but adequate for near-by portraits. The two images are slightly different in size, corrected in Photoshop by size transform.

Two ordinary cameras with the same focal length lens will do the trick - they do not even have to be from the same manufacturer. 

Any, single  ordinary camera.  

  1. If the subject stays still, stereoscopy is possible with two separated pictures from one camera.
    • Cha cha method:
      lean to the left:   take a picture
      lean to the right: take the "same" picture 
      LEFT click RIGHT click (Cha-cha)
    • Cha Cha Cha method:
      Lean left:             take a picture
      Lean right:           take the picture
      Step to the right: take the picture
      LEFT one RIGHT two STEP three (Cha-Cha-Cha)
         Three stereo separations result: 1-2 short, 2--3, longer, 1---3 longest.
         When the pictures come back, choose the best pair.
    • One - Two -- Three (Cha Cha Cha)  is a great method for beginners.

    Now jump over the hard theoretical stuff and go to here for practical advice!

Obsessional method:
slide the camera along a horizontal bar between the two pictures,
or use a variety of appliances of varying complexity for producing a rapid shift of one camera. You can invent your own methods (like the author) or get commercial systems. 

Precautions:

  1. Take one image and then move sideways for another picture.
  2. The two positions should be on the same horizontal line. 
  3. Work as quickly as possible, to avoid things (like clouds) moving between shots.
  4. Set the camera on "burst mode" and rapidly shift sideways while it is firing like a machine-gun. Surprisingly good results are obtainable on people who are not moving much.
  5. Aim at exactly the same point each time - preferably in the foreground to simplify "window effect". (Aim on a background object to avoid toe-in once you are an "expert")
  6. Never tilt the camera. (It can be corrected later while mounting the prints, but is a big pain.)
  7. If you are taking prints rather than slides, hold the camera vertical - it will be easier to view the prints later. 
  8. It is better if the two pictures have the same exposure and contrast, but not essential for 3D effect. In fact it is possible to expose one for shadows and the other for highlights and (almost) get away with it.
  9. To increase depth of field, focus one camera slightly closer than the other. When the brain fuses the stereo images it will present the correctly focussed parts and suppress the blurred parts automatically. This means stereo gives you a depth of focus bonus not supplied by flat photography.
  10. The films should be from the same batch and processed at the same time to avoid colour shifts, but you can usually get away with breaking this law too.

The next section is boring, so jump over it if you are in a rush.

Special methods

Aerial photography 

involves flying in a straight line, taking pictures on a roll of film, resulting in a sequence of many stereo pairs.

  • Stereoscopic {pictures} were taken by the Apollo astronauts as the command module orbited the moon.
  • Photogrammetry uses stereo aerial photographs for making contour maps of the earth. {Simple explanation}
A cine or TV camera moving sideways

 records a long sequence of stereo pairs. Side-ways motion produces a 3D impression in its own right, even for people with one eye.

A Passport camera

Some passport cameras have more than one lens and take two or four pictures simultaneously. These are supposed to be identical, but the lenses are separated and the pictures are actually stereo pairs, with a reduced stereo separation. In the uncut form, they can be seen in 3d using cross-eye viewing. You can even turn the "4 on" variety sideways and still get 3d portraits. By using supplementary lenses, the reduced distance between lenses allow macro stereo

Moving Subject

The camera stays still while the [subject moves], producing a 3D effect.

  • Close-up photographs in 3D are often done by keeping the camera still and rotating the subject by about 2 to 7 degrees. (The lights should rotate as well, unless flat lighting is used).
Astrophotography 

of comets or planets is possible in 3D because they move against the star background. 

  • This is a good way to find asteroids. Two pictures taken hours apart are either viewed in stereo or placed in a viewer which blinks one image and then the other, in rapid succession. If an asteroid has moved, it seems to jump backwards and forwards against the star background. (The stars are so far away, any stellar movement is not detectable in this short time). In stereo, when properly aligned along the direction of movement, the asteroid seems to lie in front of the stars. [Halley's comet] was photographed this way.
  • The {moon in stereo} results if it is photographed at the same phase in different months, because of libration in longitude and latitude. This is easier at full moon, as changes in solar co-latitude and longitude are hard to allow for during different lunations and shadow lengths change rapidly at times other than full moon. {Thurmond's stereo moon using libration}
Panorama stereoscopy

Two photographs with stereo separation taken with a panorama camera are hard to fuse when placed side by side since they are too wide. They are often presented as anaglyphs, or set up for an over and under viewer. Vertical panoramas are good for stereo viewing, if you are prepared to pan up and down. (See [viewers])

3D radiographs
  are made by:
  1. [rotating] the subject, (4 to 10 degrees)
  2. moving the X-ray tube sideways between the two exposures.
  3. using stereoscopic tubes with two foci on the anode (Toshiba, stereo parallax is limited)
  4. two separate tubes mounted together and a rapid film changer (Philips, obsolete angiography method. This causes different size images due to different FFD, which are better corrected later for good viewing)
  5. Using a C arm and rotating the tube and image intensifier around the object by about 5 to 7 degrees. The stereo sea shell was done that way.
  6. creating a [3D model from CT] or MRI data in a work station and presenting it in stereo. 

(Mirror stereo viewing works well for rotating models. Some MRI workstations are sold with liquid crystal glasses, a good viewing method for a high speed video card. Rotating pairs of angiographic MRI images are well seen with a PokeScope - or free viewing if you are up to it).

How big should the stereo shift be?

The distance between the camera positions is called the stereo base and numerous mathematical methods are available for defining it.

The simplest is the "one in thirty rule". The stereo base should be 1/30 of the distance to the nearest object.

The next section is boring for beginners, so jump over it !

Reality purists: orthosteroscopy.

  • The camera separation should equal human inter-ocular distance if the stereo impression is to be realistic and not magnified or minified. (Orthoscopic or Isomorphic stereo).
  • This means adjusting the separation individually for wide-eyed and close-set people, since not everybody has eyes 65mm apart. That sort of accuracy only happens in research projects.
  • The pictures should have the same angle of view as the taking lens of the camera, otherwise the perspective is not the same.
  • To achieve the same view angle, the transparencies should be 
         1)  viewed directly (not printed to a different size) and 
         2)  the stereoscope should have the same focal length as the camera lens. (V = F).
  • The stereoscopic effect is enhanced if it is reinforced by perspective clues. Short focus lenses are often preferred for making stereo pairs because the wide angle views they produce have enhanced perspective. This way parallax can be kept low, yet pictures still have 3D effect. But unless the same short focal length is used for the viewing lens, the angle of view is different and the pictures are not orthoscopic.
  • The two degree rule actually applies to the picture as seen in the stereo viewer, not as taken by the camera. This means the focal length of the viewer is the starting information when performing calculations. Stereo camera lenses and the maker's recommended viewer have the same focal length, or near enough. Corrections only have to be made when telephoto, extra wide angle or close up lens systems are used. The ordinary stereo photographer with his standard 3d camera is not troubled by all this since he only has one focal length to play with and no ability to vary the stereo base.
  • The two degree rule does not apply if you use "double depth masking" which is an advanced technique.

If long focus lenses are used, things get complicated. The simple case is photographing a standing person who just fits in the picture. If you use this same magnification using a telephoto lens as achieved with a standard lens, you have to move away from the subject to fit him in. For this special case, orthostereoscopy is almost achieved by increasing the stereo base by the ratio V/F, where V is the viewer focal length and F is the camera focal length. (PePax principle)
  • If you insist on including close objects, then the stereo base must be reduced when a telephoto lens is used. This is because the long lens magnifies more and so increases the parallax, compared with a standard lens.

  • The 2 degree rule does not apply to telephoto lenses, because they magnify. 
         If V is the viewer focal length and 
         F is the focal length of the camera lens:

Maximum Parallax angle = 2 V / F degrees.

  • Wide base telephoto stereo, if it photographs the same field of view as a standard lens, gives realistic stereo depth, but at the expense of reduced perspective depth.

  •  Perspective flattening is the well-known effect of telephoto lenses This means wide stereo base telephoto stereoscopy can never be truly isomorphic in a standard stereoscope. 

  • It is claimed that using the same long focal length for the viewing lens as for the telephoto taking lens restores ortho-stereoscopy, but at the expense of a tiny picture. Few people are happy looking at a stereoscopic postage stamp. Besides, you want to use the same, convenient, short focus viewer for everything.

  • Statements on the internet that telephoto stereo is not good are contested with examples by the author, especially his Amazon jungle series, which is mostly telephoto and very spectacular.

Stereo pictures are never truly isomorphic. 
  • The eyes must focus on the computer screen, but they want to change focus for objects in front of the screen ("observer space") or behind the screen ("CRT space"). (The same focus problem applies to flat prints or transparencies too).

  • The lack of congruence between focus and convergence of the eyes is worse on a small, adjacent  computer screen than on a far distant, large movie screen. Some people find watching 3d movies better than computer stereo. 

  • A magnifying lens optically takes objects to infinity, but this is NOT the same as looking at a distant movie screen. Sure the eyes focus at infinity, but the stereo parallax is even more severe, because the eyes are brought closer to the pictures in order to focus through the convex lenses. (This phenomenon is discussed further  for over-and-under viewers [here]).

  • Stationary objects in observer space must not cross the vertical edges of the computer monitor, which is interpreted strongly as a window frame. In stereo movies it does not matter so much, since people are prepared to accept things in observer space vanishing at high speed out the side of the picture. However static objects must sit behind the frame, even in movies.

  • The author's main interest is in objects which are impossible to see in 3D, unless stereo photography is used. If you are a "reality purist,"  this is not the web site for you because the orthoscopic rules are broken all the time!

Two degree rule

Stereo parallax

For computer games viewed at 18 inches (450mm) parallax is often limited to 12mm, with an absolute maximum of 20mm. Bigger separations are only used after the game has been going a while and the observer has become used to stereoscopic viewing.
Inverse tan 12/450 = 1.5 degrees. Inverse tan 20/450 = 2.5 degrees.

Note how the term "divergence" has been slipped in. Since "zero parallax" means the stereo window, it can no longer be used for the zero angular parallax of objects at infinity. So we now have total confusion of terminology depending on the book you happen to be reading:

linear parallax can only be relative. It is measured in mm or inches.

An interesting effect occurs if the observer moves away from the screen. The screen looks smaller,  but the stereo depth seems to increase. This is called "stretch." 

 


This Mingimingi seedling pokes out of the picture. Try viewing it  from different distances to see if you agree with the above - not everybody is aware of stretch!  (Some browsers will not show this properly - Internet Explorer and Opera are good. The two left pictures are X stereo and the two right pictures are U stereo)

Hyperstereoscopy

A Stereo bar and the traditional 1/30 rule

When two 35mm cameras with 35mm lenses are set up on a stereo bar and infinity is included in the picture: they should not photograph objects closer than:

N = B x 30
N = nearest object
B = stereo base

Example: Two 35mm cameras with 35mm lenses set 12cm apart:

>    N = 0.12 x 30 = 3.6 meters, around 12 feet = 4 walking paces.

Hyperfocal distance calculation for this set-up:
   f  =  F2 / Hc
       f = f number
       F = focal length of the lens
       c = acceptable circle of confusion (usually 50/1000 for 35mm format, or one thousandth of the standard focal length for the format used)
       H = Hyperfocal distance, and H is twice the nearest distance still in acceptable focus:

    f = 352 / (3600 x 2 x 50/1000)
      = 3.4
This means: 
     use f3.5 or bigger, 
     leave focus on 7 meters 
     and have nothing closer than 4 paces.
     Which means a person standing fits nicely in the horizontal frame.

Note that 35mm lenses do not produce orthostereo unless the stereoscope also has 35mm lenses.
Strictly the 1/30 rule should be modified to:
Stereo base = 1/30(V/F)
where V is the focal length of the stereoscope and F is the focal length of the camera.
Even this will not produce "orthostereo" as discussed above, but at least the stereo pair can be mounted in a standard stereo mou
nt.

 

Practical method using one camera:

The easy method is to walk sideways until there is an appreciable shift between the foreground and background objects which are to be included. 
  • Experience is the best judge of how big "appreciable" is. So beginners should  take pictures at increasing separation and decide later which are the best stereo pair.
    • You are actually aiming to have 2 degree parallax between objects on the horizon and objects near at hand. You learn to judge this without measurement or you would never get the picture taken!

    • Look at something 2 meters away and block one eye, then the other. See how much it jumps sideways compared with an object on the horizon. That is the parallax you are aiming for - so remember how it looks.

    • Your little finger held at arm's length is about 2 wide. Align the closest object with the most distant object on the left of your little finger. Step to the right until they line up on opposite sides of the finger. That is near enough to 2 parallax. (Be careful nobody becomes upset at the sight of your elevated finger...)
    • Learn what 1.2mm looks like in your 35mm camera viewfinder. In landscape format it is 1/30 of the width (landscape format). You may find a couple of scratches or part of the focusing system that is about 1.2mm apart on the view screen. Use the 2 meters versus infinity trick to find it, with a 35mm lens on the camera and a stereo base of 65mm.

    • Now use this distance to measure linear parallax in the viewfinder when using any other focal length lens or close-up system. It is far easier than calculating. 
    • In Nikkormat cameras, for example, the width of the exposure meter window is twice the desired parallax. Once you know that you can confidently use any telephoto lens or close-up extension tubes and arrange satisfactory stereoscopic viewing conditions without any calculations at all. However, you get the calculations here for those who are not satisfied with simplicity.
  • You will need to exclude objects which are too close and will interfere with the window effect (see later).
  • The camera could be aligned on one of the near-by objects as this will automatically set a good stereo window, greatly simplifying mounting the pictures later. 
  • This results in slight camera toe-in (by up to 2 degrees) which is not considered good for macro-photography but is OK for hyperstereoscopy of distant objects, where key-stone effect is not a great problem.
  • If you aim at the most distant objects, the stereo window will be in that distant plane, which is most undesirable for good stereo effect. If you do this, the pictures will have to be cut later for window. (This is in fact the obsessional way to do it, when you are more experienced, and will give the best results).

For those who are obsessional, measuring callipers, or dividers, or lines on transparent plastic can be set up at :

C = .035AV/F 

C = calliper opening
A = length of your arm
.035 = tan2
V = stereo viewer lens focal length
F = camera focal length

Then hold the callipers at arms length and measure the parallax between nearest plane and furthest plane as you move sideways - whew. 
You could always measure angles with a sextant held horizontal...

Please note that experts in stereo vision may often break these very conservative rules, at the expense of having beginners walk off in disgust when they cannot fuse the pictures. However, the experts are getting a real mind-blowing blast of depth perception and it is worth practicing just for the kicks - all achieved without drugs!

Hypostereoscopy

For extreme close-up photography (less than 30cm) most workers say the camera separation should be reduced below the 6.5cm average inter-ocular distance. This is sometimes called hypo-stereoscopy.  Most examples of excessive stereo shift are seen in macro-photography. However, there has been a recent re-think of this! A method for close-up stereo pictures, which does not involve any mathematics, is given here.

Complex formulae exist for computing stereo base. A recent versions are:
the [Bercovitz formula] and
[DiMarzio formula]. Surprisingly, the Bercovitz formula shows stereo separation for a near-by object can exceed the normal  inter-ocular distance. I was sceptical, but did an experiment, described here, and found it works. However, as they say in the advertisements for unbelievable bargains: "conditions apply". The formula also gives interesting results when applied to telephoto hyperstereosocpy, and the obsessional workers will have to check it out. Beginners, read on here.

Virtual stereo window

The aim is to have the picture behind the window frame, not in front of it, or confusion usually results.

Click here for diagram

EXAMPLE

A cocktail glass photographed in a picture frame, 
with a map in the background.
 

  1. The cocktail glass just behind the frame can be seen with both eyes.

  2. New Zealand's South Island is partially cut off by the frame in the left eye view, but the whole island can be seen with the right eye.

  3. Background portions only seen with one eye at a time have stereo failure. They are monoscopic.

Parallel eye stereo
(U)

LEFT
RIGHT
 

 

Cross-eye stereo (X)    

RIGHT  

             LEFT   

Segmental Stereo failure

Any segment visible with one eye but not the other is obviously "mono" and not "stereo". 

Suggestions to handle edge failure include:

  1. Keep the edges boring so nobody wants to look there.
  2. A "falling frame" can remove failed stereo at ground level, although it does look a bit unusual. 
  3. "Poke in the eye" technique can sometimes be used.
  4. Vignette the edges (gradually darken them). Some stereoscopes have a window built in which is so close to the eyes it is out of focus and effectively acts as a vignette edge.
  5. Avoid excessive stereo shift, which is a potent cause of segmental stereo failure.
  6. Extreme depth in the scenery is spectacular, but causes segmental stereo failure.
  7. Camera club composition rules do not apply in 3D photography. If the main object is placed on a golden third it may well end up too close to the edge and become involved in edge stereo failure

Edge stereo failure example:

Frame edges at different depths on the two sides partially corrects edge stereo failure. (X stereo).

 A "falling frame" reduces edge stereo failure more effectively than different depth frame edges.

Segmental stereo failure occurs within the frame, not just at the edges. (See here for an example which also shows excessive stereo shift and oval tree trunks from stereo magnification).

Camera Convergence

The two camera optical axes may converge on a foreground object. This ensures the virtual stereo window lies in the same plane as the foreground object converged on. Convergence greatly simplifies masking the pictures later, since the full picture frame is useable and nothing has to be cut off.
Crossing background
Keystone effect

There is no doubt our eyes converge to see close up objects, but eyes are not cameras. The eye image is formed on the inside of the eye ball, which is spherical. Camera film is flat, so the geometry is different. Quite apart from geometry, the brain makes adjustments which have been learned in infancy when the visual world and the tactile world have to agree with each other.
(e.g.: Mathematical arguments about the horopter come to nothing when the horopter is actually measured by visual physiologists, who find it makes sense visually, even if the mathematics is violated.

Projection stereo and keystone effect

Don't converge

Some workers say the two cameras should both point straight forward, to avoid both:
  1. different magnification at the edges of the pictures and 
  2. a crossing background. (!) 

Avoiding camera convergence is geometrically the same as using a shift lens to prevent converging verticals in architectural photography. The essential feature is keeping the film in the same plane for both pictures. 

A stereo camera only has one film plane and a fixed stereo shift, which means convergence technicalities do not occur in normal use. 

A stereo camera is boring to use and a pain for macro or hyper stereoscopy. However it excels for moving objects, synchronised flash and realistic stereo depth (if the stereo viewer has the same focal length as the camera lens). There is room for all tastes in this game!

If the pictures are cropped for presentation to make a window effect, then background segmental stereo failure is not cured by avoiding  convergence, although total cross-over may be prevented. A stereo camera with macro attachments may not have the windowing problem, but it is actually using convergence and does have key-stone effect... This web site tells you how to take close-up stereo with no keystone distortion, but it means using just one camera, or using moving stereo camera on a bar as if it were just one camera - another paradox!

Cutting pictures to make windows

Close-up stereo photography without camera convergence means the virtual stereo window for the uncut film frames is at infinity.
(Convergence is possible with a stereo camera if each frame is exposed separately, rotating the camera between frames.)

Personal solution to the convergence dilemma

The author used to believe (like many who write on the internet), that gentle camera convergence was good. Many pictures on this site have convergence.

After further experiments, I changed my mind. 

  • Experienced observers can handle the keystone effect and still see in stereo. 
  • Inexperienced people cannot, and there is no point making it hard for them.
  • By making pictures with and without convergence, the non-distorted versions proved superior. They were quicker to fuse and gave a better stereoscopic impression.
  • Projected stereo pairs with keystone enlarges the distortion and makes it very hard for people sitting close to the screen to fuse the images.
  • Keystone means differences in the Y plane, which must be avoided in stereo.

"Poke your eye out" effect

  • Click here for an aircraft wing poking out of the frame in anaglyph format.
  • Further discussion of anaglyph windows here.
  • Kauri seedling poking out of the frame

    "Eye poking technique" removes most of the edge stereo failure in macro stereoscopy. 
    X stereo.
    Press for Kauri tree seedling, including U stereo version.

    Computer stereo terms

    Parallax can be changed by moving the head while viewing in 3D on a computer, and a striking example is shown here

    Mounting slides or prints for a stereo viewer or projector

     

     Stereo Home Page
                                     Stereo Home Page    Stereo Home Page