Blur has been studied with isolated contours, but natural scenes are composed of occlusions at a range of depths giving rise to retinal images with broad distributions of blur. We have reported that observers are efficient at integrating wide distributions of blur with limited bias toward sharp or highly blurred elements in naturalistic, but highly controllable dead leaves patterns (Taylor & Bex, 2011 Journal of Vision 11(15) 13). To extend these findings to natural scenes, we asked observers to match the local perceived blur in a natural image to the perceived blur in a dead leaves stimulus. The dead leaves patches had blurred five different levels of Gaussian blur (? = 2, 4, 8, 16, 64 cy/image) and the observers' traced regions in the natural image whose blur matched that of the dead leaves patch on each trial. We developed a measure of local blur to relate perceived blur in natural images to the Gaussian blur in dead leaves stimuli. We found that our measure in regions traced by observers correlated well with the image blur in the dead leaves patch, indicating that the measure of perceived blur captures how blur is perceived in natural scenes. Further analyses revealed that the skew and kurtosis of blur distributions in traced regions was not significantly different from a Gaussian distribution, indicating that observers do not rely on overly sharp/blurred regions to judge perceived blur. Thus, the distribution of local image blur, rather than the global image blur or the blur of single items, determines perceived optical and image quality.
