Since their inception many years ago, all practical computer image generators have been based on producing a representation of the scene being simulated by approximating all scene features with "faces" (polygon bounded planar segments). The lack of visual fidelity in all early systems has led to continuing effort to reduce the cost per face so a greater number of faces, more closely approximating the simulated scene, could be used.

Recent developments which allow patterns to be successfully overlaid on faces promise to radically change the nature of the business. The patterns may be derived from digitized photographs, random numbers with the desired statistical characteristics, mathematical functions, or from direct input by the modeler. This provides very high scene fidelity at relatively low face counts. The user now has a choice. He can use the increased scene fidelity with conventional face capacities to address visual cuing tasks never before successfully achieved, such as nap-of-the-earth helicopter flight. Other requirements may best be met with low-face-count (and hence low-cost) systems along with the new photographic quality capability to provide more realistic cues than current high-cost systems.

This paper describes the performance of General Electric's current application of this technology (known as "cell texture") in its new COMPU-SCENE IV image generator product series.