Perceptual Organization, Figure-Ground, Attention and Saliency
Abstract
Figure and ground are often viewed as binary complements to one another, with a well defined boundary between them. A simple experiment shows otherwise: if the contour of a simple convex shape is perturbed to create a distinctive texture, it is typically the outside of the contour that provides the basis for similarity judgement, not the inside. The introduction of the appropriate task, however, can make the inside part of the contour become more salient. A similar result occurs for concave shapes, such as a C, where notions of 'inside' and 'outside' are not well defined. Here, as well as with 'holes', any proposal that directly relates figure to fixed aspects of objects fails. This leads us to propose an operational definition of 'figure'. Measures that assess similarity between shapes using a distance metric, cannot explain the above results. This leads us to suggest that there is a task-dependent bias in visual perception according to which the saliency of the two sides of a contour (inside and outside) is not the same. We suggest novel related biases such as 'near is more salient than far', 'top is more salient than bottom' and 'expansion is more salient than contraction'. We also discuss implications to visual perception; our findings seem to indicate that a frame is set in the image prior to recognition, and agree with a model in which recognition proceeds by the successive processing of convex chunks of image structures defined by this frame.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 1991
- Accession Number
- ADA259964
Entities
People
- J. B. Subirana
Organizations
- Massachusetts Institute of Technology