From “Frames Gone Wild” to “Bodacious Cha-Cha’s”

My Method of Producing Stereo Pairs from Misaligned Photos

John C. Ballou

My introduction to taking stereo photographs came about quite by accident. Airplanes have always fascinated me, and in the late ‘90s, I began to build and fly model airplanes. While flying them, I always wondered what stunning vistas must be seen from the high vantage points reached by the aircraft. The first efforts were by buying cheap (i.e. expendable) 35mm cameras that had auto-advance capability and hard-wired on-off switches. In that way, I could be assured that the camera would not go into battery-saving mode and turn off the camera just before I would trip the shutter for the most amazing shot. Fixed focus was a by-product of the low cost that was good, since I wasn’t going to take pictures of any near-field objects.

This worked pretty well, since my flying skills resulted in several cameras (and planes) giving their all, resulting in some badly over-exposed film. However, I found that there was a low hit-rate of photos that remotely resembled what I imagined the shots should look like. The choice was clear: It was time to spend a little more money, and buy a digital camera. With it, I bought a memory card that allowed me to take up to around 700 pictures. Then, things started getting more interesting. Flights that would have produced two or three interesting frames resulted in many more photos that were worth keeping. After a bit of practice, I could trip the shutter almost as soon as the camera was ready to take another shot – About 1.5-2 seconds.

At this point, another seemingly random choice that was forced on my aircraft designs resulted in opening up the stereo world to me. To minimize the cross section of the camera with respect to the direction of travel, the camera must be mounted looking sideways. That meant that if the airplane was traveling in a straight line, successive photos would show (more or less) the same subject, but with a separation in the point of view similar to a left-right camera setup.

The significance of this became obvious to me late one night as I was reviewing the shots using the “slide show” feature of the software that came with the camera. The time between shots was about 3 seconds, and the pictures just flew by. It seemed as if the slide show paused for a long time between shots, and I thought there was something wrong. After stopping the slide show, I noticed two shots that were almost identical. A light bulb lit up in my head, and I put both on my screen at the same time. Moving away from the screen, I crossed my eyes slightly, and it seemed as if the screen expanded in front of me! My excitement resulted in staying up until about 3-4AM looking through my archive for stereo pairs, feeling exhilaration when I found ones that were close.

However, there were some that seemed as if they would make great pairs, but rotation and translation of the field of view prevented viewing, and seemed to try to start a headache.

That brings us to the subject of this essay: How to combine misaligned digital images into stereo pairs.

First, you must use an appropriate program that can perform the following image manipulations:

1. Layering – puts multiple layers on top of each other so that the relative positions of elements in the image can be seen.

2. Rotation – rotates each layer individually.

3. Translation – moves the images right/left and up/down.

4. Distortion – stretches each layer by moving the corners of the image. Some programs may be able to do this by moving points inside of the image, but in my process, only the corners are used.

5. Trimming – removes parts of images that are not necessary. Note: This is different from cropping in that the size of the image is not reduced to the edges of the remaining image.

By using these manipulations, it will possible to make stereo pairs out of misaligned images in most cases. The usual cause of this is spherical distortion, which becomes an issue if the two images are taken from opposite sides of the camera field.

Here are the steps that I use for my process:

1. Create the image file that will be large enough to accommodate the right and left images side-by-side.

2. Create two layers, and place the right image in one, and the left in the other.

3. Perform rotations on one or both images to make either the foreground or the background objects line up in a vertical direction.

4. Select the boundaries of the stereo window based on the elements to be included, then trim away the excess.

5. Use the distortion tool on the corners of the remaining part of the images to line up either the foreground features (if the background was aligned in step 3), or the background features (if the foreground was aligned in step 3).

6. Place the images as desired to make a stereo card or separate them to send to a slide printer.

To illustrate this process, I will use a pair of digital images that I took while flying over a park near the shipping terminals in Oakland, CA.

Here's a pair of photos that I took of a shipping terminal in Oakland, CA. As you see, they do not completely match up. The first step is to place the pair into the same workspace, by copying and pasting them into separate layers in a new file that is more than twice the width, and around 30% higher than the size of the photos. This will allow you to move the layers around to align them, and separate them to check how well they are aligned.

Change the opacity (tranparency) of the top layer to 50% or so. Then, move both images to the center.

At this point, you must choose a feature to determine how much to rotate one or both frames to align the images. In this case, I chose the horizon, or the outline of San Francisco and the Marin Headlands. Whatever feature you choose, it is important that this feature be horizontal. You could also choose features that you assume are parallel to the line between the position from which the images were taken.

For these two images, I had to rotate both to align the horizon.

Next, move the top layer to superimpose the selected feature.

As you can see, the horizon matches up very well, but the foreground elements are not matching up in the vertical direction. Also, notice that the closer the elements, the farthere apart they are in a horizontal direction. This means that the stereo effect is present, and quite strong (after all, it is a hyper-stereo!) The next step is to distort the image to make these elements line up in a vertical direction. For this purpose, the software that I use allows you to distort the image by grabbing the corners and moving them around. It is important that you only move the corners in a vertical direction. Any horizontal movement will change the stereo effect. A friend who tried this procedure carefully lined up all the elements in his images in both vertical and horizontal directions, and asked me why the stereo effect disappeared! Well, the object here is to create a stereo pair, not to create identical images. In the case of this pair, I had to move the top left corner up a little to match up the Transamerica Pyramid, the bottom right corner down to match up the ship and the crane over it, and the bottom left corner had to move up a little to match up the crane on the left.

As you can see, these elements are now fairly well aligned. Next, it is time to place the stereo window. This is done by moving the images so that the closest elements to the camera position are aligned. The element chosen is the ship hull.

Now, you’ll notice that the images have different heights. The horizon is no longer aligned in a vertical direction. Use a resize tool to decrease the height of the tallest image to correct this problem. Do not change the horizontal size, as this will distort the stereo effect.

Next, select the area to be used in the stereo pair. The way to do this is to select the area you want, then invert the selection and erase the area outside. It should look like this.

After this step, it is important to check the alignment of foreground elements. Any adjustments necessary should be done here by distorting the image using the corner method described above. In this case, the lower left shipping containers were not quite correct, so a small adjustment was required.

Now, it’s time to change the opacity to 100% for both images, and separate them to make a stereo card!

In this case, the exposures were slightly different, so use your software to correct for brightness, contrast, and color balance. Here’s what I came up with for this pair.

In order to make stereo cards cheaply, I make my files with an aspect ratio of 3:2 (to make 4X6 prints at Target). To do this, I take the width of the pair of stereo images (with the center gutter), and multiply by 6/5. This gives me the width of the entire image, and the height is going to be 2/3 of that.

The additional space below the image allows you to place text using photoshop for your title.

The first time I went through this process, it took perhaps an hour or more to make a tolerable pair. With time and practice, I can usually make a stereo pair in about five minutes.

In most cases, it is best to ensure that your cameras are properly aligned so that these manipulations can be avoided. However, in some cases, there is not time to do so, or conditions may not allow it. In these cases, take the images anyway, and use this procedure to tame those “Frames Gone Wild”, so you can show off some “Bodacious Cha-Chas”.