Our Editor does not like to be stretched, but finds a way that is better than most.
Irecently celebrated 10 years with Sound+Image magazine — I know, it seems forever, doesn’t it, but prior to that I was hacking my way through Malaysian tea plantations, being attacked by hornets and generally having a good time, and prior to that, I was one of those UK hi-fi journalists who have such a deserved reputation for impartiality. I happened to look at my very first issue of
Sound+Image recently, and saw that my debut editorial comment piece was on the subject of aspect ratio, and in particular the horror of visiting friends and finding them watching short fat people on their screens, and apparently not even noticing. A decade on, I remain highly sensitive to aspect ratio distortions (even after spending longer in Australia and growing to understand that in some places, RSLs for example, incorrect aspect ratio appears to be mandated as an absolute commandment, whether it’s the usual short fat people or some extremely skinny ones displayed on half a screen next to exciting rolling Keno results).
Thankfully, the departure of 4:3 content has largely fixed this personal nightmare as far as televisions go, so that the last hold-out for such constraints is now the world of projection. I still sometimes get to witness a dedicated home cinema installation of extraordinary quality, only to find that 16:9 content gets partly cropped and partly stretched in order to fill a 2.35:1 (or 2.39:1 or 2.4:1) ratio screen. If I ask why, the answer is usually “because nobody likes black bars down the side”. Short fat people are preferable, apparently, though most certainly not to my eyes, even if I can see that sport, in particular, becomes a more immersive experience when filling the whole screen.
So I was most impressed recently to be shown a demonstration of a processor by Lumagen, which incorporates an idea called non-linear stretch. (It’s neither new nor unique to Lumagen, but it’s new to me.) Non-linear stretch keeps the centre of the image at about the right aspect ratio, then increases the stretch ratio further towards the edges. So at the sides, the stretch gets quite extreme. But of course our eyes register very little in the way of detail beyond the point at which we focus our attention (try reading text in the next column while keeping your eyes here). So for as long as the director keeps the action near the centre of the frame, this non-linear stretch is extremely effective at achieving the dual goals of filling the screen, yet not making people too short and fat.
It wouldn’t work for all material — you wouldn’t want to watch an Academy-ratio Hitchcock movie this way, since anything with a lot of panning will suffer almost a fish-eye effect as objects move from one side to the other. But for sport, it’s a definite success. (I should mention that Lumagen’s processors do an awful lot else besides this particular option, including HDR mapping and some apparently very impressive upscaling to UHD. We’ll return to them in future issues.)
This characteristic of vision is also being investigated in virtual reality headsets. The current crop of headsets still have, IMHO, pretty piss-poor resolution which makes their use for entertainment rather like stepping into a 480p YouTube video from 2005. But fixing that is not as easy as simply putting in higher-resolution screens, because the processing power required to generate and/or shift that high-res video according to your head movements would be monumental. You’d probably require a headset the size of a pumpkin.
But — given our inability to perceive detail where we’re not looking, the high-res screens could show low-res imagery everywhere else, and just deliver high resolution wherever our attention is focused. This idea is called foveated rendering, and clearly it will rely on eye tracking that is accurate and fast enough to get the job done pretty much as fast as your eye can dart around the image. Which is very fast — fast enough that nobody’s nailed it yet. But the goal of less processing, less power and smaller headsets has put foveated rendering very much on the VR roadmap, and Google/Daydream posted in December about their ideas for how it might be approached (see it via avhub.com.au/ fov — though be warned, it’s not light reading). Meanwhile I was delighted to experience a related idea well executed by Lumagen’s processor. Anything to get rid of those short fat people. Cheers!