Batch-processing stereograms with StereoscoPy

I have two methods of taking stereoscopic images, both of which I use regularly:

• The “left foot, right foot” method, which I can do with any camera, including that on my smart phone (I covered this method at length in my 2021 FrOSCon talk). This is the method I prefer when doing stereograms of landscapes, buildings, statues and such like, and also what works very well for posed stereo portraits.

• My Loreo stereoscopic lens, shown in the picture above attached to my camera.

Either way, I need to post-process my images to get cross-view stereograms like the one you’re seeing here.

In the former case the need is obvious: I start with two images and need to make them into one stereogram.

In the latter case, it’s perhaps less so: my stereo lens obviously already produces a stereogram, but it’s a wall-eyed one (which I’m not particularly good at viewing), and it has an area in the centre of the frame where the two images slightly overlap. I have found this area to be about 6% of the total width of the image. So that means that what I need to do, starting with the original stereo image, is this:

• Split the original image into two halves.
• Cut off 3% on the left and right of each image — on one side, that crop removes the overlap; on the other, it restores symmetry.
• Swap the sides of the image: the originally left side goes right, the right side goes left.
• For easier viewing, add a divider, and a border.

What comes in very handy here is a neat little tool: StereoscoPy is a small Python library and CLI that is helpful in batch-processing stereo images.

In combination with convert from ImageMagick, this enables me to batch-process a whole folder of stereo images into something that is much more suitable for general consumption than the original images that the Loreo lens produces.

#!/bin/bash

# Set the border/divider width, in pixels
BORDER=40

for f in *.JPG; do
    # Grab the file name, sans extension
    name=${f/.JPG/}

    # If the cross-view stereogram already exists, 
    # skip to the next original image
    if [ -e stereo/${name}-cross.jpg ]; then
       continue
    fi

    # Convert the wall-eyed stereogram foo.JPG 
    # into foo-0.jpg (left) and foo-1.jpg (right)
    convert $f -crop 50%x100% stereo/"${f%.JPG}".jpg

    # Create a cross-view image, with auto-alignment, 
    # that crops 3% off each side of the image, auto-aligns, 
    # and creates a border and divider
    StereoscoPy -x -A \
      --div $BORDER --border $BORDER \
      -C 3% 0 3% 0 \
      stereo/${name}-0.jpg stereo/${name}-1.jpg \
      stereo/${name}-cross.jpg

    # Remove the intermediate images
    rm stereo/${name}-0.jpg stereo/${name}-1.jpg
done

Maybe I’ll eventually get round to submitting a patch to StereoscoPy itself, so that the pre-processing step with convert is no longer necessary and the little script above becomes an actual one-liner. But for now this works okay for me.