Cooperativity and 3-D Representation
Abstract
Recent evidence indicates that the early stages in visual processing may be broken into several parallel streams that are specialized for the analysis of different visual attributes. A contour localization task showed that all attributes can contribute equally to border localization no particular attribute dominated position decisions. A series of experiments on transparency perception showed that transparency is analyzed rapidly (within 60 msec) and influences early levels of visual processing. We have also investigated the early stages that lead from the initial 2D representation to object recognition. Visual priming studies have been completed which suggest that object recognition begins, not with the construction of a 3-D model, but with a crude match of 2-D views to internal prototypes. The prototype that has the best match then guides the construction of an internal 3-D model. An analysis of drawing techniques has led to ideas about memory codes for the depth structure of an image, to simplified mechanisms for understanding shadows and shading and to renewed interest in 'isophot' models of shading. Visual search studies have shown that some scene features may be rapidly suppressed. For example, shadows appear to be identified early and discounted in order to allow object contours to be processed. Finally, long-term practice in visual search tasks leads to learning of both object-centered and retinotopic properties of the stimuli.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 28, 1993
- Accession Number
- ADA267180
Entities
People
- Patrick Cavanagh
Organizations
- Harvard University