Abstract
Detecting spatial patterns is a fundamental task solved by the human visual system. Two important constraints on detection performance are the variability that is found in natural scenes and the degradation of the image that occurs due to optical blurring and non-homogenous sampling of the retinal ganglion cell (RGC) mosaic across the visual field. Furthermore, most previous studies of detection performance have been conducted in the fovea with additive targets. However, image cues are different with occluding targets so these studies may not generalize well to occluding targets presented in the periphery. Here, we report eccentricity thresholds (eccentricity for 70% correct detection) for four different occluding targets presented in natural backgrounds at varying, but known, distances from the fovea. The luminance and contrast of the targets was fixed, and precise experimental control of the statistics (luminance, contrast and similarity) of the natural backgrounds was obtained using a novel method known as constrained scene sampling (Sebastian, Abrams & Geisler, submitted). Next, we describe a first-principles model, limited by known physiology of the human visual system and by the statistics of natural scenes, to compare with the pattern of observed thresholds. First, target-present and target-absent images are filtered by a modulation transfer function that approximates the optics of the human eye. Second, RGC responses are simulated by blurring and downsampling the optically-filtered image in a fashion consistent the midget RGCs at each retinal eccentricity. The model then combines luminance, pattern, and boundary information in the target region to predict detectability across the visual field. We show that a weighted combination of these three cues predicts the pattern of thresholds observed in our experiment. These results provide a characterization of the information that the human visual system is likely to be using when detecting occluding objects in the periphery.
Meeting abstract presented at VSS 2017