The Truth about 2K, 4K and the Future of Pixels
The first appearance of this article in 2008 is easily still among the most controversial articles that we have ever run, either in print or online. John is the Senior Vice President of Advanced Digital Imaging at Panavision, and is known as "the father of Genesis," a pioneering Panavision digital cinema camera. He was also responsible for developing the "Panavized" version of the Sony HDW-F900 first used by George Lucas on Star Wars Episode II, and holds numerous U.S., British, and Japanese patents in film and electronic imaging related areas. Obviously intelligent, he is also articulate, witty, and highly opinionated. In over two hours of conversation, he pulled no punches about what he thinks are the right ways and wrong ways to develop cameras. In the full-length version of the article, he also talks about the evolution of cinema, the next generation of digital imaging, and the democratization of film production via equipment rental; but here, we'll jump straight into some of the bits that elicited the strongest response. For motion picture camera sensors, the word "pixel" is a little complicated. In the old days, there was a one-to-one relationship between photosites and pixels. Any of the high-end HD cameras had 3 sensors: one red, one green and one blue photosite to create one RGB pixel. Historically, 2K and 4K referred to the output of a line array scanner scanning film, so that for each frame scanned at 4K, you wind up with four thousand red pixels, four thousand green and four thousand blue. The great perpetrators of the mythology of what I call "marketing pixels" have been RED and Dalsa. Their cameras with Bayer-pattern sensors are basically subsampled chroma cameras. In other words, they have half the number of red and blue color pixels as they do luminance, or green pixels. You have two green photo sites for every red and blue.  Note that there are twice as many green pixels as red or blue on this representation of a Bayer pattern sensor. To create a single RGB pixel, there must be an equal number of each color, so the choice is whether to discard green pixels and lose luminance detail, or to use interpolated, aliased red and blue pixels. How do you get RGB out of that? You have to interpolate the colors that are missing from surrounding pixels, creating what isn't there. You can do this extremely well, particularly if the green response is very broad -- but let's go back to scanning a film frame. The aspect ratio of a full 35mm film frame is basically 4x3. So if you have 4096 photosites across the width of the film, in red and green and blue, and 3K along the height, you multiply four by three: you'll have 12 million green photo-sites, 12 million blue photo-sites, 12 million red photo-sites. That's 36 million photosites. A 36 megapixel image is what you get from a 4K scan. But this is not what you have from a "4K" Bayer sensor. Rather than 36 megapixels, these are 8.3 million pixel sensors. You know very well that you cannot take an 8.3 million pixel sensor and create 36 million pixels from that without interpolation. You are up-converting, and there's really no value to the up-conversion. There's no new information. That's why I call these "marketing pixels." It's intentional obfuscation, because they really do nothing to improve image quality. They may improve sales volume. But they don't do anything to quality. Yet somehow the world has accepted that that's 4K. It's purely semantic. It's like saying, "I don't like my weight in pounds, so I converted to kilos. It sounds better!" You'd be amazed at how many non-technical people I meet, often producers and directors, but sometimes even cinematographers, who get fooled by that stuff. There's another fundamental problem with the Bayer sensors. In 1972, when Dr. Bryce Bayer at Kodak couldn't make sensors with enough photosites to capture full RGB, his idea to have two green pixels for every one of either red or blue pixels was brilliant. It works very well in still cameras, but with any camera with a fixed sampling structure -- in other words any CCD or CMOS camera -- you have to use an optical low pass filter to make sure that you don't create a moire pattern in the final image. If you design the optical low pass filter to satisfy the requirement of the frequency of the green samples to maintain the highest resolution, the red and blue photosites will have aliases. However, if you design the optical low pass filter to make sure that you don't get a color alias from red and blue, then you are throwing away some of the resolution from the green. So you can never get the resolution you might expect from a Bayer pattern. Someone can argue this until they are blue in the face, but we are dealing with the limitations of the physics of optics and the mathematics of sampling, and you can't escape it. PIXELS AND RESOLUTION Another problem with a message built on "marketing pixels" is that it confuses pixels and resolution. They don't have anything to do with each other. What defines the resolution, quite frankly, is the optics more than the sensor. When we released the [Panavised version of the Sony] HDW-F900, dubbed "the Star Wars camera," it was a 2/3rd inch camcorder. People asked, "Why are you doing this?" Well, because it only weighs 12 pounds, it's got a battery, there's no need for an umbilical cord, and it's got a built-in recorder just like a film magazine. Almost everyone in the industry laughed at it, but it has proved to be unbelievably successful. That camera is still renting every day with the Primo Digital lenses we designed for the 2/3 inch format, and really, you'd be hard pressed to get a better image. So you have to look at the whole system, not latch on to just one parameter and say "That's what we're gonna go for!" Everything has to work together as an imaging system. Unfortunately, one of the tragedies of digital imaging is that now we've got these ridiculous numbers games. Because so few people understand the fundamentals of imaging technology, everybody wants a number to latch on to. The numbers don't mean anything in the context of 100 years of development of film and motion picture technology, optical technology and laboratory practice. Whenever I do a presentation about digital imaging, my first question these days is, "Anybody know how many grains of silver are on a frame of film? Hands up, hands up!" Nobody ever puts their hand up. So why do we care about pixels? Because after 100 years of being comfortable with a relatively stable filmbased motion picture technology, along comes this new and disruptive digital imaging technology, and we're all clutching for some magic number that we can carry around in our heads that will define the process for us. Sorry, it doesn't work that way. It's messy and it's complicated, and lots more so today than it was in the days of film.
|
| ||||||||||||||||||||||||||||||
Reply
Like
