Purchase this article with an account.
Abhranil Das, Wilson Geisler; Understanding camouflage detection. Journal of Vision 2018;18(10):549. doi: https://doi.org/10.1167/18.10.549.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Occurrences of camouflage in nature evoke fascination and wonder in us. Less appreciated are the forces that shaped their evolution: the visual systems of their predators and prey. Indeed, having been filtered by them, camouflage specimens — no matter how ingenious — are poised near the edge of detectability, their inventiveness only testifying to the sophistication of the detection machinery that pruned even slightly less crafty variants. Using theory, computation and experiment, we have begun to investigate the mechanisms involved in detecting camouflage in nature. In particular, we consider the scenario where the camouflaging animal has exactly mimicked the background texture. In this case, the visual information usable for detection lies only at the boundary between the animal and background. We begin therefore by defining a measure of boundary mismatch: a computational measure of image discontinuities at the boundary that putatively summarizes most of this available information. We then synthesize artificial stimuli using 1/f noise as the camouflage texture (this shares the same spatial frequency properties as natural images, but lacks further structure), and assess human performance in a series of target-detection experiments. We find regular variation in the detectability of these stimuli as a function of boundary mismatch, allowing us to measure boundary-mismatch thresholds as a function of task-relevant stimulus dimensions like luminance, contrast and duration. We plan to extend this analysis to variations in the size, distance and shape of the target, and with naturalistic texture stimuli (e.g., see Portilla and Simoncelli, 2000). These ideas can also be brought to the question of engineering effective camouflage. The boundary-mismatch measure allows us to compute the best location on a background to hide against, and compare the effectiveness of different textures towards this goal. These computational results can be connected to actual detectability using the results of our psychophysical experiments.
Meeting abstract presented at VSS 2018
This PDF is available to Subscribers Only