Hyper-efficient visual detection of targets in noise

We compared human detection of visual targets in noisy images with the theoretical optimum from a matched filter. Using a small, thin target with vertically aligned markers, we obtained hyper-efficient detection. The measured detectability index was as high as 95% of the theoretical optimum, far exceeding the 50-70% typically reported for detection tasks. When the markers were removed, the detectability index degraded to 49% of the theoretical optimum, even though the placement of the target was well known to be in the center of a small panel. This result indicates the importance of inherent location uncertainty. We modeled the results using a model for human detection with location uncertainty. The uncertainty corresponded to a retinal distance of ≈7 μm, approximately the size of a single photoreceptor