Embodied Memory Allows Low Vision to Perform Like High Vision When Perceiving Events
36.451, Sunday, May 12, 2:45 - 6:45 pm, Orchid Ballroom
Jing Samantha Pan1, Geoffrey Bingham1; 1Indiana University
Introduction: In low vision, high frequency image structure is not available yielding low recognition performance. However, optic flow provides a depth map of 3D layout and motions allowing good recognition. Visual motion measurement uses low spatial frequencies available to low vision. Finally, optic flow and image structure are intrinsically related in vision because optic flow takes image to image. Pan et al (submitted) have found that optic flow information about 3D layout and progressive occlusion calibrates subsequent image structure, which then functions as embodied memory allowing identification of hidden target locations after significant delay. We now test whether this optic flow and image structure relation enables observers to recognize objects using blurry image structure that has been calibrated by optic flow. Method: Videos of eight daily events were processed with low-pass filters and thus, highly blurred. Twenty frames from each video were presented in five phases. The task was to describe each event. Phase 1: images presented one at a time. Phase 2: twenty images presented with white masks between frames (no optic flow). Phase 3: twenty images presented in sequence without masks (optic flow). Phases 4 and 5 were the same as Phase 1, except Phase 4 was immediately after Phase 3, and Phase 5 was five days later. Results: In Phases 1 and 2, the rates of correct event identification were 11% and 26%. In Phase 3, the rate increased to 88%, and dropped to 77% in Phase 4, and persisted after five days, 72% in Phase 5. With identical image structures, post-motion performance was vastly better than pre-motion performance. Conclusion: Optic flow calibrates image structures to allow accurate recognition using the blurred (nonfunctional) static images after motion ceases. Embodied memory means that low vision observers can perform better than allowed by low vision image structure alone.