caseyporn

Playing with Digital Strip Photography, Part 1

I've been playing with photos, videos, and computers again. It's getting to be quite a habit, and I must say it's quite enjoyable. This time I've been playing with "strip photography", but I thought I would start with the story and get to the details later. Read on for more.

(This is a more reasoned followup to this previous article.)

So I was investigating - again - whether there were any digital cameras that were, for want of better words, weird, quirky, strange, or eccentric. This was related to my digital lomo train-of-thought, and also my rediscovery of Lensbabies. So I did what anyone does, and started doing some Google searches.

One of the first pages I came across was this blog entry which pointed to this page in the DPReview forums, and showed some amazingly unexpected photos coming out of a cheap digital camera. The straight fans of a moving wind turbine had be turned into strange curves, but the rest of the picture was normal.

I recognised what was going on pretty quickly. One of the first links I ever posted on the web was to a high-resolution digital camera made out of a scanner. Since the scanner took a noticeable amount of time to complete a picture, it had the generally-unwanted side-effect of taking the same kind of weirdo pictures as were posted on DPReview, where moving things were very distorted.

The consensus on DPReview, by the way, was the same: the camera in question (also reviewed by Dan's Data, the greatest of all tech review sites) is a dodgy digital camera that is a linear scanner: that is, it collects information off the CCD line by line (horizontal lines, top-to-bottom in this case) instead of all at once. And so, when things are moving, you get weirdo results.

Now, the Digitrex DSC2100 is an old camera (some seem to still be available on ebay) and not a very good one, but it seemed that I had found a geniunely "quirky" digicam - hurrah! - and one that I could, potentially, own. But why should I? Maybe I could just fake it, instead?

Armed with my trusty Conwa video camera, the current pinnacle of low-tech high-tech tomfoolery, I wandered out my backdoor in my tracksuit pants and shot a short video. The camera was on a tripod, and recorded me waving my arms around in the air.

To simulate the effect of the quirky-cam, the idea was to take one horizontal line out of each frame of video, moving down one line at a time. This is easy enough to code in Python, thanks mainly to the Python Imaging Library (I first used VirtualDub to save a sequence of images from the video). There are lots of practical issues to resolve, though: taking exactly one line from each frame would mean that, for a 480-line picture (TV-quality) it would need 480 frames of video, which is 16 seconds. I wanted to be able to grab one "action" in video, and use that as the source, which meant grabbing more than one line from each frame. Additionally, since it's hard to grab fractional parts of lines (well, impossible, really), I had to build in some intermediate upscaling as well (thanks to Euclid for the greatest-common-divisor algorithm).

The end result was a bit of code that took an arbitrary-length sequence of images and took chunks from each of them, and composited them back together. Hang on a second, that rings a bell. Just recently I made Multimatic, which is a flexible system for making Lomo-like composite images from digital sources. Now, I didn't use Multimatic for this task, but instead wrote just a little bit of custom code instead. It wasn't too hard, and was more efficient.

The results, as you may have already seen, were pretty interesting. Below is a picture made from video of me sweeping my arms from above my head to by my sides. The video was scanned from top to bottom:

As there were not very many frames, more than one line is taken from each one. You can see the "jaggies" quite clearly. If I had taken 16 seconds to move my arms, I would have been able to take one line from each frame, and the result would be very smooth. Being able to scale the images makes it easy to experiment without precise timing.

In the picture below, I rotated one arm, and then the other. The image was scanned from left to right:

So as the scan moves from left to right, it captured first my right hand, heading up over my head, and then captured my body as it had turned to the left. At the centre-point, both physically and temporally, I turned to the right and rotated my left hand.

This next one was from the same arm-rotating sequence as before, but only the first side (right hand). You can see what a difference the time-sequence makes to the result:

For the next photo, I did a little arm-waving dance. The aim was to get fast movement and hands in multiple spots. I didn't have any clear idea as to how it would turn out - in this example, it was scanned top to bottom:

Oh, and I had to try the "photofinish" style of photo, where a single line is taken from the middle of each photo. For this, I moved my hand very slowly past the camera, close enough that my whole hand never fit in the frame, making pictures like this:

When you stick them all together, then re-stretch it to match the original width, you get things like this (NB: the camera was hand held, not on a tripod):

From this very short experiment, some things were made pretty clear to me. First, this is fun! Second, the images that can be produced make for interesting artifacts, and there is potential for a whole range of different effects. Third, doing this digitally allows for a lot more flexibility than doing it using specialist equipment. It's much more flexible even than buying a quirky digital camera to do it. The only downside is the low resolution of my capture device - namely, a shonky little video camera. A specialist camera - or, indeed, a proper video camera - will offer much better quality output. But hey, at least I can have a play.

Now, in a display of exactly the wrong type of thinking for a research scientist, once I'd had a play I tried to find out what had been done before. It took a lot of hit-and-miss searching, trying to find the right words that other people had used to describe the same type of stuff. Here's the summary of what I found:

• The original analog technique is called strip photography, since it involves exposing only a strip of film at any one time. It's what is used in so-called photo finish cameras, which expose a picture on the finish line.
  
• I'm far from the first person to do it digitally. As well as Matthias Wandel, who was already known to me, there are many many people who have experimented with this. First, it was adapting scanning technology for the purpose: one of the progenitors of this area seems to be Andrey Danidhazy, whose site and images are worth checking out. Also, I found this blog entry about Michael Golembewski, who makes scanner-camera pictures. His site is interesting, also. Other links: Java-based video slit-camera, list of slit-scan works, the sCaNoN modified camera.
  
• Other people have experimented with doing this with video, too. I saw a couple of works at World Without End, an exhibition at the Australian Centre for the Moving Image in Melbourne, including one by Daniel Crooks who has played with similar ideas in his work.
  

So I'm not breaking any ground here, really, but it has been a lot of fun.

The aims and results of this and Multimatic are worth thinking about together. Multimatic (and, by analogy, Lomo Sampler-style photography) is concerned with compositing images large enough to stand by themselves. There are only a few (four or eight) images, and the time-series is short too - just a couple of seconds. The intention is for each image to be seen by itself, and then as part of a sequence. In this kind of strip-photography I've been doing (it's not REALLY strip photography, not even really digital strip photography, since it does a little more and a little less), the contribution from each image is very small, and cannot be viewed as meaningful by itself. In this case, the emphasis is on the composite image to the exclusion of the individual image. There's a point - I don't know where, but I would guess it would be around ten images - at which this occurs. Multimatic lives on one side, strip-photography on the other.

What next? Well, once you've gone to digital video capture, there's a whole word of possibilities. The big thing is, you have a lot more data to choose from. Now, there's some graphs coming up, but don't get scared.

Traditional strip photography, whether implemented on film or using a scanner or digital camera, takes a picture over time, one line at a time, horizontally or vertically. You can think of this as having all the pictures stacked on top of each other, then choosing the pictures like this:

Nice smooth planes, aligned with either the x or the y axis (both directions are shown here, since they're trivial to do too). This is because that's all the data they have access to - they are the only photos you can take. (There's one more option - photofinish cameras where the lens doesn't move capture a vertical plane.) Once you have all the data, though, you can do more. For instance, not particularly interesting, but you can tilt the plane so that you're taking a diagonal line each time, which might suit your intended effect better:

But why stick with a plane at all? You're just describing an arbitrary function that maps from an (x, y) co-ordinate (one specific pixel) to a specific time that you want to take the pixel from. So you could get wacky:

Even this example assumes that your only choice for each pixel is how far back in time you'd like to look - the frame of reference is fixed. You could just as easily choose pixels from anywhere in the data, but the meaning of this gets a bit more questionable.

Anyway, I haven't done any of that stuff (well, I have sort of - my code allows arbitrary gradient planes, whereas traditional strip photography has a fixed gradient of one), but maybe I will, who knows?

Well, that's all for now. Thanks! If you're reading this and you'd like to collaborate or something, do drop me a line.

Posted by Casey at May 16, 2005 06:15 PM