Perceiving Human Motion in Synthesized Images
Joseph D. Anderson
Jessica K. Hodgins
THE GOAL OF special-effects animators is to create on the computer portions of an event that can be intercut or composited with other portions of the event that were shot in live action, such that the computer-generated portions of the event are indistinguishable from the live-action portions. To achieve this goal is no small accomplishment, yet most of the feature films we see contain computer animation, and a good portion of that work passes our notice. The artists and technicians who work to create special effects have gotten very good at their jobs, yet they generally proceed by way of “making and matching” (see Gombrich, 1960, p. 29), and our theoretical understanding lags considerably behind their accomplishment.
If the problem is to match computer-generated images with photographed images, then the most obvious question is, what are the differences between the two images? One rudimentary difference is that there are more grains of silver halide in the picture area of the film than there are dots in the matrix of a standard computer image, giving greater picture resolution and a greater range in gray scale to the photograph. But perhaps much more important are those differences that accrue to the two types of images because of the way they are made. The computer-generated image is constructed; the film image is recorded. The question becomes: What are the differences between a constructed image and a recorded image?
To achieve a computer image, objects and events are often modeled mathematically. The resulting image can be realistic because the world we know seems to follow physical laws and therefore to lend itself to mathematical description. In theory, the relationship between an object in the world and an object on the screen could be just as exact if created by mathematics as if created by the mechanical action of light rays—if it were not for the problem of specificity. When an event occurs in the world, the event itself is unique, and it produces a disturbance in the optic array that is also unique. However, the disturbance in the array is not a picture, not a “copy” of the event. The particular relationship that exists between the event and the corresponding disturbance in the optic array is one of specificity. The information in the optic array specifies the event. As James J. Gibson noted, “There are different kinds of disturbances for different kinds of events” (1979, p. 102). It is this problem of specificity with which the animator must deal.
For example, when a camera records the explosion of an automobile, a common event in the movies, the explosion occurs according to the laws of physics, the gases expand and combust in irregular patterns, and the car separates into millions of fragments of unpredictable shape and size that move on complex unpredictable paths. The camera records every detail of the event that can be seen from its point of view, and when the film is later projected on a screen, the viewer finds authenticity in the consistency of