Abstract
Objective: The visual field generally appears veridical and undistorted in spite of variations in retinal projection and our poor sensitivity to spatial form in crowded peripheral vision. Here I measure sensitivity to spatial distortion to examine the accuracy of visual representation. Methods: Spatial distortions were generated within one quadrant of large natural scenes, pink noise or phase-randomized natural scenes. The transition from undistorted to distorted regions was smoothed with a Gaussian window. The period of distortion was varied between 0.25–16 c/deg using band-pass filtered noise, newly generated each trial, to control the spatial displacement of remapped pixels. The magnitude of distortion was under the control of a staircase to generate distortion Modulation Transfer Functions (dMTFs). The observer was required to indicate which quadrant contained the distortion in a 4AFC task with feedback. Results: Sensitivity to spatial distortion decreased with eccentricity (measured out to 8 deg) and required deformations as high as 2 deg. For natural scenes, dMTFs were band-pass with a peak for distortions over 1 c/deg. Sensitivity to distortions within pink noise or phase-randomized natural scenes was much higher and dMTFs for these stimuli were low-pass. Conclusions: The fact that sensitivity is higher for random phase images suggests surprisingly that the presence of edges and contours in real scenes masks the presence of distortions. The 1 c/deg peak in dMTFs suggests that contour-finding processes prefer orientation changes within around 1 deg. Overall, humans are extremely insensitive to spatial distortions in natural scenes, suggesting that, like the perception of blur, the appearance of veridical visual space is entirely an illusion.