Abstract
We move our eyes to view different parts of our surroundings. But how do we integrate these views from one fixation to the next? Several studies have shown that we are, in some ways, surprisingly bad at piecing together a picture of the world; for example we are blind even to considerable changes in the visual display when these changes occur during saccades. Other studies, though, reveal that quite precise visual spatial information survives the eye movement. To explain this performance, we note that there are two methods the brain could use to combine views from different fixations. One method, called updating by oculomotor physiologists, uses motor or proprioceptive information about eye motion to account for the resulting shifts of the retinal image. The other method, called mosaicking by computer scientists who use it for example in radio telescopy, relies on visual information alone, using common features in separate snapshots to glue the snapshots together in a unified picture. Both updating and mosaicking are error-prone in different ways. Our model of transsaccadic integration uses both methods, combining them in a way that optimizes the expected alignment of successive images. This optimal combination, which depends on the statistical properties of the visual, proprioceptive and motor signals and of the outside world, explains some of the strengths and flaws in human transsaccadic integration.