The perception of multiple objects: A parallel, distributed processing approach

To what extent can information distributed across the visual field be processed in parallel? This thesis reports on the development of a neurally-inspired model of two-dimensional shape recognition capable of identifying multiple objects presented simultaneously on its retina. The model, called MORSEL, has four components: (1) a set of processing modules that analyze objects along various attribute dimensions; (2) a system that constructs a consistent interpretation of the perceptual data provided by these modules; (3) an attentional system that guides the efforts of the modules; and (4) a visual short-term memory that holds object descriptions. MORSEL has been trained to recognize letters and words in arbitrary retinal locations. Following this training, it is also able to recognize several items simultaneously. However, MORSEL shows capacity limitations of two sorts: first, cross talk among items can ensue, and second, some location information is lost in the course of processing. These limitations restrict the nature and amount of information that may pass through the system without error, and appear to match human limitations. MORSEL does quite well in accounting for a broad spectrum of psychological data, including perceptual errors that arise when several objects appear simultaneously in the visual field, facilitatory effects of context and redundant information, and attentional phenomena. The thesis also reports on a series of experiments which show that, as predicted by MORSEL, repeated instances of an object fail to be detected when focal attention is restricted.