Abstract
In the literature on visual stability, most studies focus on the problem of relating two retinal images, one before and one after a saccade. The more general problem of achieving visual stability over a number of saccades or for a freely moving observer requires a different solution. The answer must lie in one of two categories: visual stability could result from the generation of an unchanging representation of the scene despite head and eye movements, or it could reflect an ability to predict the sensory consequences of actions (including saccades and head movements) without relying on a stable, Cartesian representation of the scene. The former approach entails 3D coordinate transformations as an observer moves their head and eyes; the latter requires a large storage capacity. Examination of the storage-based approach leads to a change in perspective for other visual problems. For example, under this model, retinal flow should not be decomposed into rotational and translational flow. That might be useful for extracting 3D structure and observer movement in a stable coordinate frame but it is not appropriate if the purpose of retinal flow analysis is to determine the trajectory along an expected path of images. As observers carry out tasks, their sensory+motivational state moves along an expected path. Task-dependent paths are built up gradually through evolution and development. The experience of visual instability would arise only when the expectation is confounded. Results such as the apparent stability of an expanding virtual room (Glennerster et al 2006, Current Biology) can be explained within a storage-based framework but are more difficult to account for on the basis of Cartesian scene representation. Overall, if we are to make progress in understanding visual stability, we must be clear what the problem is that needs to be solved.
Meeting abstract presented at VSS 2014