Showing posts with label upscaling. Show all posts
Showing posts with label upscaling. Show all posts

Friday, August 3, 2007

The Truth about Upconverting DVD Players

HD and related technologies aren't that well understood by the general public. We've all seen the hype about "true HD" displays and about "upconverting/HD compatible" DVD players, and not many people care about or understand what these things mean to their viewing experience. I had a friend recommend that I get a 1080i compatible DVD player instead of getting BD or HD-DVD right now while I wait out the format wars. He claimed his DVD picture was as good as HDTV. I've read advice that says that for fixed panel displays, upconverting DVD players match the native pixel count of the display, and thus provide a better picture. I'm here to say this:
all of the above is bull. Whether we're talking about interlaced vs progressive, upscaling DVD or not, its all fancy marketing terms to get you to think you can get a better picture. The truth is, both of these rely on real-time interpolation, and its a question of which device has the best algorithm.

What is interpolation?
It is any method of guessing at information you don't have. For example, we all know that an interlaced signal shows you all the even lines of one frame, and then all the odd lines of the next. For a CRT, the phosphorescence lifetime (the time it stays bright) of the pixel plus the phenomenon of human visual perception called the persistence of vision effectively blur the picture to where we don't notice we are seeing half the data for any given frame. But for any fixed resolution panel display (LCD, PDP) this wouldn't be the case. These each have a native resolution (usually 1920x1080 or 1280x720 pixels), and what you see on the screen is in this resolution, ALWAYS. For example, I have a 1920x1080 display. When I view broadcast 1080i HD from my cable box, somewhere along the path from where the cable enters the cable box to where image gets rendered on the screen, the picture has been converted to 1080p. Period. The cable box could do the conversion, or the TV could. In my case, the TV does it. It performs a deinterlace. It separates the even and odd lines of the pictures, then guess at the missing pixels, then shows you the two frames back to back. What about native 720p content from my cable box? In this case, my cable box is set to upconvert this to a 1080i signal (so my TV isn't always switching modes from channel to channel). How does it do this? Interpolation. It takes the pixels it has, spaces them further apart (with unknown pixels in the middle), then guesses at the unknowns. The 1080i signal gets to the TV, where it is deinterlaced (again by interpolation) to give me my 1080p display. DVDs? Same thing. A regular DVD player outputs 480p. This gets upconverted by my TV to display at 1920x1080.

Wrong Focus in the Consumer Market
So while people are worried about upconverting DVD players giving a better picture, they don't realize that their TVs do the same scaling themselves. What they should be worried about is HOW each device does its interpolation. There are various algorithms for interpolation (zero-order, bilinear, bicubic, all sorts of proprietary adaptive ones based on identifying motion, etc). Consumers should want to know the details of how each one works, and a side by side comparison. This comes up for my regular old DVDs: do I let the PS3 playing the DVD do the scaling/upconverting to 1080p, or do I let my Sony TV? The answer is "whichever one is better at it" but at this point it is impossible to find comparisons such as this available to consumers. You might think that with them both being Sony, they should be the same, but the truth is there is no way of knowing other than side by side comparison. I suspect the PS3 has more raw DSP capability, and my tests confirm that getting the PS3 to do the upconversion gives a visibly better PQ.

The bottom line is this: there is a fundamental idea in information theory that says that no algorithm can recover 100% of missing information for all inputs. So a particular algorithm might do really well for some inputs, but MUST fair horribly for some others. It is all about what devices have the best algorithms for the type of signals we feed them, namely moving pictures. And this is something you have to decide with your own two eyes.