With an eye to low vision: optic flow enables perception despite image blur.

PURPOSE From static blurry images, it is difficult to perceive objects because high spatial frequency details are filtered out. However, in the context of events (defined as objects in motion), motion generates optic flow, which provides a depth map of 3D layout and allows good event perception. Visual motion measurement uses low spatial frequencies that remain available in blurry images, making events perceivable. Optic flow and image structure are intrinsically related in vision because optic flow takes one image to the next. Optic flow is powerful in specifying depth structures and it calibrates the degraded image structure; image structure is persistent and it preserves events perceived with ongoing motion, after it stops. Might optic flow and image structures interact and allow events to be perceived despite poor quality images? The answer to this question has implications for event perception with low vision. METHODS Twenty blurry images depicting each of eight daily events were used as stimuli. Ten normally sighted participants perceived the stimuli and described the events in five ordered conditions: (1) when single frames were presented, (2) when all frames were presented with motion masks, (3) when all frames were presented without motion masks, (4) when single frames were presented, and (5) when single frames were presented 5 days later. RESULTS With blurry static images alone, participants were unable to identify events. Events were perceptible when the blurred images were played in sequence, making motion-generated information available. Subsequently, when given the original blurry static images again, post-motion performance was vastly superior to the pre-motion performance. Furthermore, the high rate of recognition persisted after 5 days. CONCLUSIONS Optic flow calibrates low-quality image structure to allow accurate event perception during and after motion. This implies that low-vision observers should perform much better than allowed by weakened image structure information alone.