Abstract
Despite considerable variability in the visual appearance of natural scenes, they share many statistical regularities. Firstly, natural scenes are similar in their photometric properties as they share a unique distribution of luminance intensity variations known as the 1/f⊠ amplitude spectrum (⊠ ≈ 1). Secondly, natural scenes are similar in their geometric properties as they each contain a similar density of structure across spatial scales—a property that makes them fractal (e.g. in how the branching pattern of a tree is similar irrespective of scale). Since the visual system has evolved in a natural environment, it is likely that it is tuned to both its photometric and geometric properties—but to what extent? Is it critically reliant on photometric characteristics which can change dramatically depending on the illumination of a scene? Or is it preferentially tuned to geometry which remains stable irrespective of illumination? Here we use both psychophysics and fMRI to measure perceptual sensitivity (4AFC “odd one out” task) and BOLD responses in visual areas V1–V4 (N = 10) to three different stimulus image types—greyscale, thresholded, and edges. While each image type shares the same geometric properties, their photometric properties differ dramatically. If the visual system is preferentially tuned to natural geometry, we should observe virtually no difference in the pattern of activity across image type conditions—which is exactly what we find. Both sensitivity and BOLD activity show a characteristic dependency on the geometric properties of an image, peaking for stimuli which had the most natural geometry across all image types despite large differences in their photometric properties. This suggests that both behaviourally and physiologically, the visual system is critically reliant on the geometrical, fractal, structure of natural scenes—a property of which remains stable irrespective of scene illumination.