Abstract
PURPOSE Computer animation simulates motion through rapid presentation of spatially offset images. As well as temporal aliasing, the discrete nature of individual frames leads to an artifact we term ‘ghosting’, perceived as multiple copies of image features. Although only low frame rates are required for perception of motion, ghosting persists beyond the flicker fusion frequency and limits of current computer displays. We consider the basis for such artifacts and the use of synthetic motion blur to eliminate them.
METHODS A model for scene sampling, display, and early visual processing was developed and tested using psychophysical experiments on human subjects viewing either a true motion stimulus (rotating drum) or a computer simulation, with or without artificial motion blur. In an object recognition task, subjects were asked which of two alternative face-sketches moved past an aperture. In a motion detection task, subjects were asked which of two-alternative sequences of (2 or more) frames contained motion.
RESULTS Our model shows that ghosting results from the slow response of photoreceptors (temporal low-pass filtering). Neural images of un-blurred features presented on one frame persist into subsequent frames where they reappear at new locations. Without motion blur, object recognition was possible in animations at velocities far beyond those where performance fell to random levels for true motion. Un-aliased motion energy in animated scenes is not affected by artificial motion blur and we found that performance in a motion detection task was not significantly affected by its use.
CONCLUSIONS Artificial motion blur (temporal anti-aliasing) in computer animation of moving scenes at moderate frame rates eliminates ghosting to produce more realistic simulation of motion. Eye movements may confound this solution, so complete elimination of such artifacts would require frame rates over 1000 Hz, or motion blur based on real-time eye tracking.