ereidell@media-lab.MEDIA.MIT.EDU (Evan A. Reidell) (04/12/91)
Rendering using stochastic sampling with jittered subpixel locations introduces noise into computer animations that makes each frame appear "alive" and flickering. Pretty neat, but it still doesn't have that "film" look, it just looks a bit noisy and fuzzy. Richard Greenberg of R/Greenberg Associates hinted that he had some interesting algorithms which faked "the look" for use with his stuff, but he didn't say what they were. probably top-quality trade secret, eh? can anyone "in the know" think of good SIGGRAPH papers that do their best to describe why film stock looks the way it does, and how that might be synthesized as a filter through which CGI rendering takes place? how about scientific photography papers, are there any good books or documents that discuss the mathematics of that "film" look? why we can recognize the difference between a black-and-white movie made in the 50s and a black-and-white movie made today, within seconds? how can we port that look to computer graphics? Any pointers or discussion welcome, especially about the "look" of visual displays, video, film and animation. Evan A. Reidell (ereidell@media-lab.media.mit.edu)
mccool@dgp.toronto.edu (Michael McCool) (04/12/91)
ereidell@media-lab.MEDIA.MIT.EDU (Evan A. Reidell) writes: >can anyone "in the know" think of good SIGGRAPH papers that do their >best to describe why film stock looks the way it does, and how that >might be synthesized as a filter through which CGI rendering takes place? Don't know of any Siggraph papers on the subject, but I think a good start would be: 1) Simulation of Grain Noise 2) Simulation of Logarithmic Intensity Response. Film grain is Poisson, wheras stochastic ray-tracing uses jittered or Poisson-disk sampling, which have different characteristics. In particular, they introduce less low-frequency noise, and hence let the Monte-Carlo integration technique converge with fewer samples. To get the "photographic" look, you need to sample completely at random. You could probably do it by randomly sampling an image, but make sure you interpolate between pixel values. You should use a different pattern for each frame. Then average the samples down. I'm not sure off the top of my head if you can just add noise, it may be image intensity-dependent. There's probably some way to figure this out and just add the noise without having to simulate the resampling. The apparent amplitude of the noise interacts with the intensity response. Film also has a logarithmic intensity response, and you would have to simulate the correct "gamma" for film. This is fairly straightforward, just map the colours. Most image processing books talk about film response and then ignore it for the rest of the book. I think Wintz has a discussion of film response. Also see Barrett and Swindell, "Radiological Imaging", for a really good discussion of this topic. In general, I think the curve for film bends *opposite* TV monitor response, at least for negative film. This reference talks about both noise and response. Other effects are possible with film, such as "Solarization"... like in (too many) music videos (Eurythmics comes to mind). Flash some light on the film just as it finishes developing, and you get and edge-enhancement effect, and washing out of colour. This should be pretty easy to simulate with a little image processing. In the shadow of the sun, Michael McCool@dgp.toronto.edu, Dynamic Graphics Project, University of Toronto.
rthomson@mesa.dsd.es.com (Rich Thomson) (04/13/91)
In article <5644@media-lab.MEDIA.MIT.EDU> ereidell@media-lab.media.mit.edu (Evan A. Reidell) writes: >[...] why we can >recognize the difference between a black-and-white movie made in the 50s >and a black-and-white movie made today, within seconds? When I worked in a movie theatre I learned that different film stock degrades over time in different ways. Technicolor is nice because it keeps its colors "bright" without much decay. Other film stocks can fade to a purple color, even when the original movie was black and white. These may be the differences you're seeing between old B&W movies and new ones. Film stock has changed considerably in the past 40 years, I imagine. -- Rich -- ``Read my MIPS -- no new VAXes!!'' -- George Bush after sniffing freon Disclaimer: I speak for myself, except as noted. UUCP: ...!uunet!dsd.es.com!rthomson Rich Thomson ARPA: rthomson@dsd.es.com PEXt Programmer
spworley@athena.mit.edu (Spaceman Spiff) (04/13/91)
I would think that the color sensitivity of film would play a bigger role in the "film" look than the finite grain size. The film color gamut is typically much larger than any video display. A video display also has a nearly flat color sensitivity, wheras file does not have a uniform sensitivity to color. Delicate aquas are one example that comes to mind- they're definately less vibrant in film as opposed to pinks. Another thought about simulating the film "look" is a more detailed camera model. Certainly depth-of-field and motion blur are known and used, but how about subtler effects like change of focal character from the center of the focal plane to the edge? Or the film shutter- exactly how much time should be integrated over to characterize the light entering the camera? What about the beginning and end of the integration where the shutter is half-way covering the lens? I don't know how much effect these would have on a computer generated image, but if you're after a film-like look, these should probably be looked into. Scratches and dirt might also be interesting. Wow- try to model a hair or piece of lint that moves up the screen, and finally disappears. We've ALL seen this at the theatre! Good luck, Evan, on your quest. -Steve --------------------------------------------------------------------------- Steve Worley spworley@athena.mit.edu ---------------------------------------------------------------------------
rhbartel@watcgl.waterloo.edu (Richard Bartels) (04/13/91)
The discussion reminds me of a topic we had in my freshman anthropology class oodles of years ago. Whenever a new technology arrived during the development of civilization, it was invariably packaged to look like old technology until the innovators learned how to exploit its distinctive potential and the consumers learned to accept its strange novelty. Examples included the fact that the earliest recorded pottery was invariably decorated with incisions that made it look like basketry. -Richard