Abstract
How is our visual representation of natural scenes constructed with constantly-moving eyes? We introduce a new change blindness paradigm, combining Blackmore et al's moving-image paradigm (1995) and Sampanes, Tseng, and Bridgeman's progressive transformation paradigm (in press). A picture jumps repeatedly to random locations on a screen, with a simultaneous change made to a local region of the image during the jump, until the image is finally “morphed” into another one in 13 steps. To avoid misalignments of parts of the old and the new images, a checkerboard of white squares is superimposed on the image while the remaining isolated visible squares are being substituted. When gist is altered between the pictures ( e.g. a girl holding a glass of water vs. the same girl holding an eclair), changes are always detected. However, when gist stays the same between the two pictures (e.g. one dumpster vs. a different dumpster), participants fail to detect the the change 70% of the time. On the trials where they do detect changes, their detection rates accumulate slightly within the first four transforming steps, suggesting that a visual representation does take some time to be built. These responses then reach an asymptotic level and stay evenly distributed throughout the remaining 8 transforming steps. In other words, change detection does not get easier as changes accumulate, suggesting that each image is compared with its immediate predecessor. Together these results confirm the idea that only a few perceptual details and a gist can survive a saccadic eye movement or image jump and be retained to aid conscious detection. In addition, these details accumulate initially, and are refreshed after each sample. Therefore, our experience of a stable and detailed visual world is a construction from only a few previously fixated details, and a gist.