Pixel Aspect Ratio is one big modern digital hassle. You'd think pixels would be square and, by and large, you would be correct. On modern computer monitors and all HD televisions, pixels are square. It turns out that the media and movies we record with out camcorders and playback on those devices often do not use square pixels, however. DVD-Video uses pixels that are slightly shorter than they are wide. Well, SD DVDs do, anyhow, and so do the original SD digital video cameras. Widescreen DVDs, on the other hand, are encoded using pixels that are taller than they are wide! And many modern digital cameras, even ones that cost hundreds of thousands of dollars, use even taller pixels. The strange part about all of this is that it's really a very old technological relic that predates digital media by many decades. And while you might think the non-square pixels might go away in the near future as we all move to square pixel monitors, my bet is that it won't, because it's really a pretty ingenious trick. But I'm getting ahead of myself.

First, let's look at DVD-Video, since this reveals the issue very clearly. NTSC DVD-Video frames have a resolution of 720x480 pixels. More precisely, DVD-Video frames, in both 4:3 standard and 16:9 widescreen versions, have a resolution of 720x480. So how is it possible that the same number of pixels can form a frame that is both squarish (4:3) and widescreen rectangular (16:9)? The answer is, of course, that the pixels are actually stretched by changing the Pixel Aspect Ratio (PAR), which thereby changes the aspect ratio of the entire frame. You can see pixel aspect ratio display errors most clearly with perfect spheres, globes, circles and grids, but once you know to look for it, you'll see it in human faces too, especially down at your local sports bar when they get new televisions and haven't yet figured out how to configure them correctly. (Click the image thumbnails to see larger versions of the included illustrations.)

You'll also discover the wonders of PAR displays and conversion when editing video. Video from your SD camcorder will come into your computer with a pixel aspect ratio of 1:0.92, but your computer display uses a square PAR (1:1) and pictures taken with your digital still camera will also use a square pixel format. Different editing software handles this in different ways, but many editors and most compositors like to see the actual pixels they are working with, so when you are editing, your SD camcorder footage might appear a bit stretched in your preview window if you look closely. Some apps have the ability to simulate different displays as well, so, for example, in Sony Vegas, you can toggle the preview window to show you what your video will look like on TV when displayed correctly (right click the Preview window). In other words, spheres and globes will appear perfectly round when Vegas simulates the view for the project you are working on. For SD 4:3 DVD-Video, this means your display will be fairly square and a little narrower, since the 720x480 0.92 PAR pixels need to be thinned a little. This is handled online on video hosting sites like YouTube by displaying 640x480 square pixels. For widescreen 16:9 DVD-Video, the screen aspect needs to be stretch a bit from the 720x480 1.21 PAR pixels. This comes out to 640x360 square pixels on some onine video hosting sites.

It is sometimes suggested that this is a result of "old" NTSC CRT tube televisions utilizing rectangular pixels (presumably 0.92 PAR), but this is only sort of true. The original analog NTSC spec doesn't really have digital pixels, only lines, and I'm a little fuzzy on why DV camcorders from last century and DVD-Video discs went with non-square pixels. The processing technique is still used today, however, even in many of the really high end formats as utilized by Panasonic's VariCams and Sony's CineAlta models (as used by George Lucas for Star Wars). The reason why non-square pixels are used in these situations is purely economical. Take, for example, the 1080i broadcast format, which is the highest resolution format used today at 1920x1080 pixels (square, of course). That's 2.1 megapixels of image sensor, data processing, bandwidth and storage. If you shot 1440x1080 pixels with a PAR of 1.33:1 instead, you only need to work with 1.6 megapixels of data. That's a 25% savings, which is very significant.

Furthermore, this is not a new technique and is still very useful even if the data rate, processing and storage issues get handled (as they surely will). Think about how movie cameras work (the kind that take film). For many decades now, 35mm film has been the standard format for big-budget Hollywood productions of all sorts. But when you look at a piece of 35mm film, you'll immediately notice that it is, well, relatively square. In fact, the frame aspect ratio looks a lot like plain old television with a 4:3 aspect ratio. And yet we know the movies we see in the theater are much wider, sometimes 16:9, but often even wider than that and sometimes epically wider still. If I can perform a quick conversion of these ratios to ":1", 4:3 becomes 1.33, 16:9 becomes 1.78 and most movies hit the screen at 2.4 (or 2.39). Ben Hur (1959) was projected in theaters at 2.76:1 (although it was shot in a 70mm film format). So, how do you get a 2.41 aspect ratio frame onto a 1.37:1 aspect ratio piece of 35mm film? Well, you use an aspherical lens to squash the image sideways in an anamorphic process and then, back in the theater, you use a correspondingly aspherical lens to stretch the image back out on the screen. As you can see, this is a very old technique and a very useful one (even if it can be mindbogglingly confusing).

Which brings us full circle back to today and into the future, where I think anamorphic shooting and non-square pixels will continue to be used even after all camcorders convert to full-frame 1920x1080p60 (which is happening as we speak). First, it's obvious that the 16:9 aspect ratio (1.78:1) of this shooting format and the standardized broadcast format AND the standardized widescreen television manufacturing format is inadequate for cinematic applications at 2.4:1. The solution: optically anamorphic lenses, of course and a corresponding digitally anamorphic process in editing (and maybe distribution). On the production end, this means anamorphic lenses to squash a 2.4:1 image into a 16:9 aspect ratio shooting format. Digitally, it's easy to handle, but even if we eventually see real cinematic displays from companies like Philips (with square pixels at 2582x1080?) or imaging chips and camcorders that shoot more and more pixels, there's always going to be someone who wants to go wider. Lawrence of Arabia 2010, anyone?

References:
* The Widescreen Museum, tremendously cool, organic and fun site
* Cinematic Television, from Philips
* Shooting 16:9 Widescreen, holy cow, an article I wrote for my personal Website more than six years ago!