3D CANYONS
3D MACROS
3D GALLERY
3D MOUNTAINS
3D GUIDE
MICROSCOPY
3D MICROSCOPY
WILDLIFE
TRIP REPORTS
TECH NOTES
RESEARCH
FUN & GAMES
ABOUT SITE
HOME
 

GETTING STARTED IN DIGITAL TWO-IMAGE 3D PHOTOGRAPHY

John Hart

Program in Atmospheric and Oceanic Sciences

University of Colorado

Boulder, CO 80302

hart@tack.colorado.edu

nimbus.colorado.edu/hart/science.htm

Dec. 21, 2004

Download a .PDF Version of this Article

This article provides an overview of how to get started in digital 3D photography.  No messing around, I get right to what I have found to work the best, with links to further discussion and other resources.

WHAT'S SO GREAT ABOUT DIGITAL 3D?

You get to do stereo photography without all the hassles of messing with film chips that must be carefully handled and tediously mounted.  From start to finish, you can do everything digitally.  This means a quicker path to the end result - stereoscopic images that can be seen by family and friends, who will probably be astounded, even blown away.

WHAT IS TWO-IMAGE 3D PHOTOGRAPHY?

You use two small digital cameras, mounted side by side, and triggered together, to take a left and a right image of each scene.  These are then processed digitally using  free software in order to produce a "stereo pair".  Here is an example from a slot canyon.  The left image is on the left and the right image is on the right.

HOW DO I VIEW 3D IMAGES?

Digital pairs can be viewed on computer monitor (or two), in print form, or via digital projection.  There are three common formats:  parallel, cross-eyed, and anaglyph.

Parallel Pair:  Left image on left, Right image on right.

Cross-eyed Pair:  Left image on right, Right image on left.  Try

Full Color Anaglyph                                                Half-Color Anaglyph:

The "half color" version sacrifices color fidelity in favor of less ghosting and color rivalry that can make viewing  a full color anaglyph somewhat painful for particular colors, like white.

Inexpensive viewing aids are available.  Prism viewers help people to look at parallel pairs.  Red-Cyan filter glasses are necessary to look at "anaglyphs".

Prism glasses (top) Pokescope (below) for viewing parallel pairs.       Anaglyph glasses.  (alternate source)

WHAT IS CRITICAL ABOUT THE CAMERA SETUP FOR 3D PHOTOGRAPHY?

1)  The lenses of the two cameras should point straight ahead, and be aligned vertically by mounting on a parallel base.  A wide angle zoom lens setting works best to begin with, although normal and telephoto settings are useful too.

2)  The minimum separation of the lenses should approach that of human eyes (i.e. it should be about 2.5 to 3 inches, or so).  Actually it should be about 1/30'th of the distance to the nearest object in an image (a rough guide).  A 3 inch base separation will give good stereo effect for near-subjects about 8 or more feet away.  For close-ups the separation should be smaller, for very distant scenes it should be wider.  Such situations are called hypo- and hyper-stereo respectively.

3)  Some method of firing the cameras simultaneously, so that the shutters are "synchronized", is required for all but still-life 3D photography.

WHAT CAMERA SYSTEM WORKS THE BEST?

As of November 2004, there is one system that, in my opinion, is the best for digital 3D photography.

The three critical categories are satisfied by 1)  appropriate mounting, 2) using small consumer or prosumer digital cameras, 3) using a special remote control that starts the cameras off in-sync, and then tracks their performance and tells you when the sync error is unacceptable (at which point you need to power down and then restart the cameras).

The rig uses two Sony V1 digital cameras (the newly available Sony V3's will also work, but are more expensive).  These are mounted on a "Z-bar" in which the left camera is upside down.  This permits a closest separation of just a little over 3 inches.

"Finished"  Z-bracket holding 2 V1's and the LANC control.  Stereo base can be quickly adjusted:  ~3.1 to 5.7 inches.

Prototype or DIY bracket (front and rear views).

The "LANC Sheperd" twin camera controller by Rob Crockett.

The cameras are controlled by a micro-processor based camera-powered unit called a LANC Shepherd.  This talks to the cameras through their accessory ports so that no modifications to the cameras are necessary.  The smaller white side-button turns the cameras on.  About 75% of the time they come up "in-sync".  If they don't, they must be powered down and restarted (using the same little white button, which toggles the camera power).  Sync is indicated on the small LCD screen of the controller.  For example, the above image indicates that the shutters will fire within about 0.88 milliseconds of each other, adequate for just about any action shot.  Sync drifts slowly, but you have a good readout of what it should be at all times.  The cameras will typically remain in sync (i.e. within 1/500'th second or so) for several minutes.  The large white button is the exposure/focus preset (half pressed) and trigger (fully pressed).  Other controls run the zoom lens and allow for time-lapse photography.

In using this controller to sync the cameras, both cameras should be set up the same (ex.  A - mode, flash off, same focus mode, etc.), and each scene should be pre-focused using the half-press of the trigger button.  If you just push the trigger all the way, depending on where the cameras' lenses were previously set, one may take a little longer to focus, which is required before the camera will fire.

Cameras are available from numerous companies on the internet.  I have had good luck with B and H Photo.  

The controller is available from Rob Crockett (the originator), Berezin, and Pokescope (at least).

The complete LANC Sheperd users guide is also available online.

HOW DO I HOLD THE CAMERAS?

There are various ways to mount the cameras for 3D.  You want them pointing parallel.  In the inverted position (as above), the minimum horizontal separation is about 3.1".  This is substantially smaller than with non-inverted mounting (~4.3").  Vertical inverted mounting leads to the smallest minimum separation, but this format is not terribly good for projection because to get to the 4:3 (W:H) aspect ratio most all projectors use a lot of cropping has to be done.  The SPM site has some images of other cameras (not V1's) and various mounting configurations.   Rob Crockett has a description of the inverted-vertical fixed-separation V1 mounting .

I prefer the inverted horizontal mounting, because it has the smallest minimum separation (for the horizontal format).  A good mount should have an adjustable base.  This is one reason why two independent cameras are so useful for 3D photography - you can do hyper-stereophotography with a lens separation substantially greater than that of human vision (about 2.75").   Here is how constructed the one pictured above.  Make the camera mounting plates (right picture above) out of 1/8 x 1.5 inch aluminum angle (Home Depot, Lowes, or True- Value).  The back lip is for alignment and the plate should be attached to the camera via its tripod socket (1/4-20 thread).  Another 1/4-20 screw is glued into the plate with epoxy (it's the one sticking up in the picture).  Both screws should be countersunk into the aluminum so the screw heads do not protrude.  It may be necessary to sand down the screw heads a bit.  Then the mounting plate will ride flat against the camera and against the Z-bar (left picture above).  It is necessary to glue the plate onto the camera (or use double sided sticky tape), to keep it from twisting.  The back edge of the mounting plate should be aligned with the back of the camera.  This keeps the cameras pointed parallel when attaching the cameras to the bracket.

 

                                                                      A Z_Bracket that holds the two cameras.                                      The V1 mounting plate.

The bracket is made from 1/8 by 2" angle aluminum (also from Home Depot, Lowes, or True-Value).  The back edges are left untouched, but the fronts and sides are cut down with a hacksaw (or bandsaw).  The slots are cut with a coping saw.  The only critical piece is the 3/4" diameter column, with a central 1/4" hole (non-critical alignment), but which should be 2.56" tall (for the Sony V1).  Contact the author for this piece if you cannot fabricate it yourself.  Below is a shot of the bracket parts (including a black tripod quick-release, optional).  The two Z-plates are set parallel on a good surface while the bolt attaching both to the cylinder is tightened securely.  A bit of lock-tite or epoxy helps here to maintain alignment over time.  A simple small hunk of angle holds the controller (via plastic-weld epoxy, but don't cover up the access screw holes).

There are probably easier ways to do this (see all the other configuration referenced above), but I wanted something very robust and repeatable for outdoor adventure sports photography in difficult environments.  Commercial brackets are available from Pokescope  but are of a different, somewhat less effective IMO, design that gives a larger value of the minimum separation for horizontal mounting.

 

HOW DO I PROCESS THE IMAGES?

This is the most interesting part, and why digital 3D is time-effective and affordable, and fun to do.  It is relatively trivial to manipulate the images in software.  Computer manipulation is far easier than dealing with film chips, with problematic dust, scratches, etc., which has always made slide and print-film based stereoscopy rather tedious and difficult.

FIRST, DOWNLOAD THE REMARKABLE FREEWARE PROGRAM STEREOPHOTO MAKER (PC's only).  Here is the link.   Having this in hand, here is how you process a bunch of images from the two Sony V1's.  BE SURE TO GET THE LATEST VERSION, AT LEAST stphmkre222 (version 2.22).

The following instructions, in particular step 3, assumes you are using a Z-bracket like that shown above, and that your left camera is inverted.  If your cameras are NOT INVERTED, you can skip step 3 below, but load your left camera's images into the LeftC folder (for consistency with the write-up that follows).

1)  Set up a directory with folders like this:

2)  Now download the cameras' jpeg images.  Dump the left camera to the Left folder, and transfer the pictures in the right camera to the Right folder.  You end up with something like this (where the camera's files increase numerically):

3)  After you have downloaded your camera files, the first task is to re-orient the images from the left camera.  Just out of the left camera, these are inverted because of the upside-down mounting of this camera on the Z-bar.  Start up StereoPhoto Maker (hereafter, SPM).  Go to the File Menu and Choose Open Image file list .

The left side of the window that opens allows you select the drive and file folder you wish to operate on.  We choose the 3Dprocess main file and the Left folder as shown below.  The file list then appears on the right side of the window:

The View control on the toolbar offers several alternatives.  Select Thumbnails .

At this point all the inverted files appear:

To tell subsequent operations that these images are inverted, click Exif on the top toolbar and then select Set the orientation tag 'Invert':

This operation enters data into the jpeg file header that instructs SPM to invert the left images in subsequent operations.

NOTE:  Alternatively, if you have Photoshop, you can write a very simple ACTION to do an automatic batch rotation of the Left directory.  This Action simply opens (OPEN) a file, rotates the canvas 180 degrees (ROTATE CANVAS, 180), and saves (SAVE) the file as a top-quality (12) jpeg (or however you want to save it), then CLOSE.  Use FILE, AUTOMATE, BATCH commands to run the action.  Photoshop's mirror and invert functions are clean.  If you do it this way, skip part 3 as above.  One advantage of doing it this way is that all other programs (like Photoshop, Paintshop, for example) will open the left files as they should be.

4)  The next task is to inspect the Left and the Right Directories side by side to make sure that the total number of files in each are identical, and that the Left and Right components of an image pair line up.  What can happen is that perhaps only one camera may fire (this does occur on rare occasions, if, for example, one has not focussed during the pre-set).  Then all the pairs after that will not line up.  If we try to build stereos out of these everything past this mismatching pair will be totally screwed up. 

a) Start SPM and do OpenImageFileList as before.  Note:  As a shortcut, this operation is on the SPM main toolbar.  From the File List, select the Left directory.

b) Close SPM (but not the List Directory Window).

c) Open SPM again, and do another OpenImageFileList operation.  But now choose the Right directory.

NOTE (Dec. 23, 2005):  An upcoming version of SPM will have an Instance command on the OpenFileList menu:

Instead of steps b) and c) above, you can Open another Instance to set up the right window alongside the left window.  Then you can use U or D keys to scroll both windows simultaneously up and down, respectively.  This makes the process of comparing long sequences of images from the two cameras, as described below, much easier.  Another toolbar option, View, allows you to specify the image size.  This is also useful in comparing left and right sequences.  Make it extra-large if there are subtle details that vary from picture to picture:

In the end, this should leave you with two windows, showing the left (normal orientation) images and the right (normal too, of course) images, respectively.  You can drag the corners of these windows to resize them, and drag the top bars to reposition them.  Minimize the file menu by dragging the right edge.   Use these operations to position the windows for comfortable comparison of the L and R images.  Perhaps you may wish to set them up vertically in a single column.  Then you can use a Pokescope viewer, say, to look at pairs in 3D.  Such a setup is shown below:

Perhaps a better setup to look at a lot of pairs in to include more in each window:

Now, for example, notice that the top two rows of pictures in both the left and right panels are the same.  However, the 5'th pictures (as you read a book) in the two panels are different.  The Left folder has a tree, while the Right folder has a roof.  In addition, you may notice that the Right folder has more files.  Evidently the left camera didn't fire when making the roof picture, or maybe the right camera fired accidentally.  To re-align the entire series of L-R image pairs, the Right folder's roof image should be deleted (or cut and saved if you really like the roof picture).

To delete the roof, simply select it and delete it (or cut and save it elsewhere):

Scrolling back down to where you were, you should now see that the image sets line up.

Continuing down from top to bottom in the directories you should check that all images pairs are appropriate.

In the above example, the difference (roof vs. tree) was obvious.  If you have a sequence of nearly identical image pairs, it may be harder to see mismatches.  Here the use of a Pokeskope (or similar) on a single or double column view of the L and R pairs may be helpful.

 

5)  Now, since the numbers that came out of the camera files are somewhat arbitrary, and probably are very different (as in the example of section 4), we should rename the Left and Right folders' files with a more useful title.  Go back to SPM and choose Multi Rename from the File menu.  Working on the Right folder (in the Look in: box), do a Full Rename .  Type in the name you want (here I have chosen BatchTest_ , and make sure the last box in Full Rename says _R.jpg .  With the boxes set as shown above, click on Rename All Files

Renaming the Right Folder files.

Now repeat this Rename operation on the Left folder.  To do this, first select the Left directory in the Look in: box and change the last box to _L.jpg .  Finally, after the renaming is complete, go to the resulting Left and Right folders and cut all the files to the main 3DProcess (here) directory.  Ctrl-A selects all files, Ctrl-X cuts them, and then after returning to the main directory, 3Dprocess, Ctrl-V moves them over.   In the end the main 3Dprocess directory should look like this:

There is a bunch of equally named left and right independent images, nicely lined up and ready to format into 3D stereo pairs!  As a check, for each number there should be both a _L and a _R file.

6)  Make up a batch of Parallel Pairs.   Again choose Multi Conversion from the SPM File Menu.

Set up the conversion as shown.  We are operating on the 3DProcess directory, starting with the BatchTest_001_L file (click on it to put it in the File name: box).  We are going from Independent(L/R) (Input file Type) to Side-by-Side (Parallel, Output File Type), and placing the resulting files into the Output Folder as shown. The Output Folder is the Parallel one we set up in the beginning.    Image quality is high (99, the default) and we will write JPG files (obviously there are many options here).  One little addition is that while converting all these files we are going to make a slight +20 pixel vertical Position shift to the left image to better align the pairs (obviously you can make a lot of adjustments).  This +20 shift was decided on after checking a few pairs individually, perhaps after the first pass you do with all this.   Once you have set up for the conversion, as shown, click Convert All Files.

It is at this point that I usually do a backup.  The parallel pairs residing in the hard disk  E:/3Dprocess/Parallel are saved to a CD-R or DVD.  I like to use Verbatim DVD-R media, which is based on their "metal AZO" technology that is claimed (and at least partially verified by accelerated tests) to have a 100 year archival life.  All such discs should be stored in a dark dry cool environment, and should be handled as infrequently as possible.

7)   Now we are ready to look at some pairs!   From the File Menu chose Open Stereo Image and Open the first file in the Parallel folder (which is obviously a Side-by-Side format).

You should get a nice parallel pair that you can view with a Pokescope or Prism Glasses (or free-view if you can).  Resize the window by dragging the lower right corner to make viewing comfortable. 

 

Alternate viewing methods are obtained via the Stereo menu:

Of these, perhaps the Color Anaglyph (or Half-Color, via the Color Anaglyph submenu) is the most useful to quickly check through a large number of files, or to look for all-important registration errors.  Entering the half-color mode, we can see that the vertical parts of this example image do not line up too well. 

 

NOTE:  At this point, I usually get out my anaglyph glasses and scan through the whole directory (hitting the space bar moves you forward, while backspace moves you backwards up through the Parallel directory).  I write down the numbers (like 10, 17, etc) of each parallel pair I really want to get ready for prime time.  For example, suppose I like BatchTest_010, along with some others.  I then select these by Ctrl-clicking on the filenames in the Parallel folder, and copy them to the Parallel Registered folder in the 3Dprocess (or whatever you named it) directory.   Now I am ready to fine tune these selected images.

8)   Registering for the Stereo Window:  Open a stereo-image from the Parallel Registered (or Parallel) folder that you want to fine tune.  Now hit the  K  key to enter the alignment mode:

The alignment window can be viewed WITHOUT the anaglyph glasses.  Basically you use the controls in the Easy Adjustment box to correct for any registration errors and to "set the stereo window".  In the above example this means correcting for the fact that the helmets are skewed vertically.  One uses the V. position slider to do this.  To set the stereo window we slide the red (left) frame horizontally until the red and cyan features in the nearest object coincide.  This is called setting the nearest object "at the window" (see the above link for details).  One uses the H. position slider and/or the associated arrow buttons to do this.  There are other corrections in size and rotational orientation that can also be done.  For example, if your zooms don't quite match up, resize the offending L (red) or R (cyan) frame by using Image Size.  If the Left (red) image is twisted, unlink the Rotation operation and adjust the Left images rotation using the degree commands.

After performing the operations (moving the red (left) image 61 pixels up, and 50 pixels to the left, in this example ) we get a pretty well-registered pair.  You will notice the difference if you look at the above image (vs. the un-registered version) with your anaglyph glasses.   Note how easy this is!  Compare this with even the thought of taking film chips out of their slide mounts, pushing them around with needles while looking under a magnifying glass, correcting for viewing parallax in this process, focusing on a ruler background then back on the images, and finally then trying to remount the film chips precisely into heavy and expensive glass mounts without getting a single speck of dust on them.  Wow, digital is great!

After you've done the alignments, hit OK.  This returns you to the main screen.  You then Crop off any bits that don't have information in common (e.g. in the anaglyph view, a red border or cyan border that stands alone).  The Crop function is in the Edit box, or just click the crop symbol on the main toolbar (the little square thing noted by the arrow below).  Drag the crop frame around the part you want to keep, and hit Enter.  Crop options (such as controlling the aspect ratio) are in the Edit, Crop Options box.

In this trumped up example (above), the arrows in the image show areas where a rotation correction has left information that is incongruent and not in both frames.  Such areas must be cropped off (as shown).

Saving your stereo pair:

Finally, you can save the registered pair in whatever stereo format you are displaying in the main frame after clicking OK on the alignment menu.   Always save at least the side-by-side version.  To save a side-by-side version, make sure it is being displayed that way (via the Stereo tool command) on the main SPM screen before typing S, or entering SPM's File menu to get to the Save Stereo Image dialog box (above).  Other formats (like anaglyph, say) can be saved by re-displaying the image that way, and then re-entering the Save Stereo Image box.

After you have saved your work you can hit the spacebar in SPM to move on to the next file in your Parallel Registered folder.

 

WHAT CAN I DO WITH ALL THESE STEREOS?

Now that you have a bunch of well registered stereo pairs, what are some options for showing them?

1)  Make some prints on an inkjet printer.

    a)  Make small (7" wide) parallel pair prints and send them to your friends along with prism glasses.

    b)  Make larger (8 1/2 by 11, say) anaglyph or half-color anaglyph prints and send them to your friends along with paper or rigid frame anaglyph glasses.

2)   Use a computer display.

    a)  View parallel pairs using the Pokescope.  View cross-eyed pairs without glasses, if you can.  Maybe invest in shutterglasses.  Anaglyphs, which are easily made in different forms by SPM (using the Stereo menu), can also be viewed on a computer screen when using appropriate anaglyph glasses.

    b)  Submit some work to a 3D gallery or join a Stereo Club that has online folios.

    c)   Use two monitors for a really high quality computer presentation of stereo pairs.

    d)  Go for the max and get a 3D No-Glasses display:

Autostereo 

3)  Project your work.

    a)  Use two digital projectors to create a passive polarized projection "slide show".

   b)  Take your work to a National Stereographic Association (NSA) convention and let them project it for you.

4)  Perhaps move a step in the other direction, and use your digital files to create 35mm slides that can be looked at in a hand viewer.   Digital files can be converted to transparencies at a photo-lab that has a digital film recorder.  Or, you can make your own if you have an old film camera and a good high-resolution computer monitor.  Some people believe high quality hand viewers (which do give complete separation of right and left images) provide a very good way to look at stereo pairs.  You probably have to mechanically remount the slides (which is what we wanted to avoid by doing digital in the first place - but the numbers may be small).

However you proceed, it's a good sport.  Have fun out there.

Return to the 3D GUIDE

Home