Segmenting Multiple Closed Contours

(joint work with Karvel Thornber and Lance Williams)

Abstract

Using a saliency measure based on the global property of contour closure , we have been able to reliably segment out salient contours bounding unknown objects given an edge image. The measure also incorporates the local Gestalt principles of proximity of edges and smooth continuity of contours that previous methods have exploited. Our measure incorporates closure by finding the eigensolution associated with a stochastic process that models the distribution of various contours passing through edges in the given scene. In this paper, we derive an explicit relationship between our saliency measure, which is the relative likelihood that smooth closed contours pass through a given edge, and the corresponding components of the eigensolution. Then we present a segmentation algorithm that utilizes the saliency measure to extract out multiple closed contours by finding strongly-connected components on an induced graph. Results are presented on several real images where we also compare the performance of our saliency measure with those that use only local properties of contours. With our approach we are able to segment real images with an average of about 10 secs per object on a general-purpose workstation.

• Introduction
• The Affinity Matrix
• Saliency Measure
• Segmentation
• Results
  • Fruits on Concrete
  • Fruits on Grass
  • Stones on pavement