Using Image Analogies for Rapid Generation of Terrain for Simulations, 07-R9698Printer Friendly Version
Inclusive Dates: 04/01/07 10/01/08
Background - While daytime, summer imagery for terrain databases for simulations are widely available, there is a shortage of nighttime and seasonal (fall, winter) terrain imagery. Furthermore, most nighttime terrain is simulated using "light points," i.e., glowing dots placed to indicate man-made structures. While this suggests nighttime lighting, it in fact is not very realistic, as it fails to show large lit areas such as baseball diamonds or well-lit "ribbons" of road. For this project, a potential method for quickly and easily generating terrain imagery by manipulating summertime imagery was investigated.
Approach - Image Analogies (IA), a method for constructing generalized filters from training images, was used to generate terrain imagery. The IA algorithm derives a filter by comparing two images and determining the modifications that must be made to the first "unfiltered" training image to produce the second "filtered" training image. IA algorithms are then used to apply the derived filter to subsequent images to produce the same basic effect. The IA process has previously been demonstrated to work well on relatively simple filters, but previous implementations were too slow to be practical for large amounts of imagery. An attempt to speed up the algorithm was made by performing the bulk of processing on the Graphics Processing Unit (GPU) on commercially-available computer graphics cards.
Accomplishments - A partial IA implementation was developed that works on very small images. However, due to GPU memory management limitations, the GPU implementation was no faster than previous implementations which used the Central Processing Unit (CPU). As the GPU and CPU become more tightly integrated and GPU programming libraries improve, the GPU IA approach will become feasible.