At Home with John and Debbie

Here’s a panoramic photograph. It’s a bit strange because some of the lines aren’t straight, or are curved in ways you wouldn’t even expect from a wide-angle lens, or in most panoramic photos you’ve seen before.

Here’s another one, where everything’s gone mad.

Here are some more.

These pictures were created using a technique which I’ve called “Slivr” (formerly Videorama, but that title was overloaded).
The technique involves “scanning” a scene, by taking a video which pans across it, then converting the resulting video file into a wide still picture by taking a narrow slice from each frame and glueing these together into one image. In effect we are modelling what happened in analogue “slit scan” panoramic cameras, which would draw a strip of film past a slit, while moving the lens across the scene.

I’ve written some more on the background of strip scan photography — you may have come from there.

This is a tutorial on how to do it.

Software

This tutorial uses two free applications for Windows.

• VirtualDub, an excellent video processing tool for Windows.
• ImageMagick, a useful command line image processing tool for many platforms, including Windows.

Install these before trying out these instructions.

Taking the video

The movie mode on most digital camera is suitable for experimenting with this technique. Hold the camera in portrait orientation (i.e. on its side) and point it at the left hand side of the scene you want to capture. Start filming, and slowly and evenly pan to the right. Later on, you can experiment with less even movements, but for now try to keep it even.

(Aside: we take the video in portrait orientation because the width of each frame is not important to us — we’ll be throwing most of it away. The taller the frames are, the better, because that gives us the greatest vertical resolution in the finished picture.)

I recommend starting with about 15 seconds of footage for every 45 degrees you film: at 15 frames per second (typical on a stills camera’s movie mode), this gives you 225 frames. With experience, you’ll get a feel for the right speed to use for a subject.

Processing the video into strips

Our next job is to split the video into image files, each one a tall, narrow slice. VirtualDub can do this in one step.

Make a folder for this project, and save the video to this folder. I’m using “C:\pano\office\”.

Launch VirtualDub, and open your video with “File-> Open video file…”.

You’ll see the first frame of your video, twice:

The left hand image is the video you’ve loaded. The right hand image is the video as it will be output when VirtualDub has processed it. You can “scrub” through the video frames using the slider at the bottom.

The first thing we need to do is rotate the video so that it pans left to right, as we filmed it.

Choose “Video->Filters”. If this is greyed out, then choose “Video->Full processing mode” first. This brings up a list of filters — it should be empty for now.

Click “Add”, and select “Rotate” from the list of filters, and click OK.

The rotation you need will depend on how you held the camera.

Click OK on all the dialogues until you’re back to the main window. You may need to move the scrub slider to see the result, which should be that the right hand image is the right way up.

Next, we need to crop the frames. There is no dedicated filter to do this, but VirtualDub allows cropping as a side-effect of each filter.

Choose “Video->Filters” again, and click on the rotate filter you just set up.

Use the Filter input cropping dialogue to set the crop. This crop happens before the rotation, so make a wide, narrow bar. Either drag the top and bottom of the picture, or use the “Y1 Offset” and “Y2 Offset” fields. Either way, you can work out how big your strips are by subtracting “Y1 Offset” and “Y2 Offset” from the original height of the frame. E.g. if your original is 320×240, then Y1=119 and Y2=119 leaves you with a 2 pixel strip, which is a good value to choose as a starting point.

The position, as well as the width, of the crop will affect the final result. Experiment.

Click as many “OK”s as you need to get back to the main screen. The right hand image should now be tall and thin, as the slices we want to create will be.

We’re almost finished with VirtualDub — all that remains is to tell it to output our frames. Choose “File->Save image sequence…”

In the save dialogue, set “Minimum number of digits in name” to enough digits that the first and last frame filenames are the same length. This ensures that when the slices are joined into a panorama, they go in the right order.

Leaving the directory field blank causes VirtualDub to save to the same directory as the original video file. TARGA is a suitable output format, for no other reason that that it’s the default.

Click OK, and watch your project directory fill up with TARGA files (the TARGA icon will vary, depending on what graphics applications you have installed).

Joining the slices

To join the slices, we use the “convert” command from ImageMagick.

Open an MSDOS prompt, “CD” to your project directory, and issue the command “convert *.tga +append pano.jpg”

(Aside: look how much easier command lines are to document than GUIs)

Finished!

That’s it. View pano.jpg in your favourite viewer. Mine looks like this (only bigger):

You might find that yours isn’t quite what you wanted. Maybe it’s wider or narrower than you wanted. Maybe some part of it isn’t the shape you expected. Experiment with wider or narrower slices, taken from different parts of the frame, or resize your image in your favourite image editor. Have fun, and consider posting your best work to Flickr with the “slivr” tag.