Abstract

We address the problem of camouflaging a 3D objectfrom the many viewpoints that one might see it from. Givenphotographs of an object’s surroundings, we produce a sur-face texture that will make the object difficult for a human todetect. To do this, we introduce several background match-ing algorithms that attempt to make the object look likewhatever is behind it. Of course, it is impossible to exactlymatch the background from every possible viewpoint. Thusour models are forced to make trade-offs between differentperceptual factors, such as the conspicuousness of the oc-clusion boundaries and the amount of texture distortion. Weuse experiments with human subjects to evaluate the effec-tiveness of these models for the task of camouflaging a cube, finding that they significantly outperform na??ve strategies.