Abstract
One of the most fundamental natural visual tasks is the detection of specific target objects in the environments that surround us. It has long been known that the properties of the background have strong effects on target detectability. The most well-known properties are the luminance, contrast, and similarity of the background to the target. In a previous study, we showed that these properties have highly lawful effects on detection in natural backgrounds, and that human detection performance is strongly linked to the statistics of natural scenes (Sebastian et al., under review). However, there is another important factor affecting detection in natural backgrounds that has received little or no attention in the masking literature, which has been largely concerned with detection in simple backgrounds. Namely, in natural backgrounds the properties of the background often vary under the target, and hence the target is not uniformly masked by the background (i.e., some regions of the target are masked/occluded more than others). We refer to this effect as submasking. To begin studying this factor we measured detection thresholds in simple independent-noise backgrounds where the luminance, contrast, or texture orientation varied under the target. In each case, the backgrounds were designed so that a classic matched template (MT) detector performed equally well whether or not the background varied under the target. This classic matched template detector is the optimal detector for backgrounds that do not vary under the target. However, for backgrounds that vary under the target, we show that the optimal detector weights each pixel location by its estimated reliability. We found that human performance tracks the performance of the reliability-weighted matched template (RWMT) detector. We demonstrate that humans make use of this same principle when detecting targets in natural backgrounds.
Meeting abstract presented at VSS 2017