EXTRACTING PERCEPTUAL STRUCTURE IN DOT PATTERNS.

The authors describe an approach to structural segmentation of dot patterns such that the results are in agreement with segmentation performed by humans. Voronoi neighborhoods are used to represent the geometric structure in the dot pattern. A set of experts then, in parallel, look for structural components such as borders, interiors, and curves. Expertise is expressed in terms of the properties of and interaction among Voronoi neighborhoods. The interaction is accomplished through a relaxation, constraint-propagation process. The results of these experts have errors due to the lack of local evidence for the global role of a dot. Each result is corrected so that it agrees with the results of other experts and it provides locally smooth borders. Except for occasional gaps in the borders the union of the corrected results represents a good approximation to the perceptual structure in the dot pattern. Connected component analysis is carried out to identify the gaps. The gaps are filled to close the component contours, ensuring local border smoothness.