Abstract
An ultimate goal of vision science is to understand performance under natural conditions. We describe a direct experimental approach for identifying and quantifying the factors that affect detection performance in natural scenes. A large collection of calibrated natural images is divided into millions of background patches that are then sorted into narrow bins along the dimensions of interest. In the present study, each bin represents a particular (narrow range of) mean luminance, contrast, and similarity (phase-independent correlation of the background to the target). Next, detection thresholds are measured parametrically for a sparse subset of bins spanning the entire space. The psychometric function for each bin is measured by randomly sampling background patches from that bin, without replacement. Finally, we analyze the residual variation of the background patches within each bin for other factors that strongly correlate with the measured performance. We find that in the typical natural image amplitude thresholds vary by approximately two orders of magnitude. Further, threshold amplitude is a linear function of mean luminance (Weber’s law for luminance), threshold power is a linear function of background contrast power (Weber’s law for contrast), and threshold amplitude increases linearly with similarity once above a base level of similarity. We also find that the three dimensions combine systematically, in a fashion consistent with a mixture of separable and additive interactions. Finally, we identified another dimension, “contrast-contrast”, that explains some of the residual variance in the thresholds: all else being equal, thresholds tend to be lower with higher variation of contrast within a patch. We argue that the results may form the foundation for a general model of detection in natural scenes. We also argue that this direct experimental approach should be applicable to other natural tasks, if a sufficiently large set of natural stimuli can be obtained.
Meeting abstract presented at VSS 2015