Authors: [tex2html_wrap4130]J. Elder, S.W. Zucker
Investigator username: zucker
Subcategory: computer vision
The boundaries of objects in a scene project as one-dimensional contours in a visual image, while object surfaces project as two-dimensional regions partitioned by these contours. In a complex scene these contours and regions are fragmented by occlusion and shadows, which greatly complicates the task of perceptual grouping, i.e. the task of organizing the fragments to support object recognition. Given the topological duality between region and boundary one would expect the visual system to jointly exploit both region and boundary information to group these fragments. While this is a standard approach in computer vision, the human visual system appears to follow a different strategy. We report here results of visual search experiments which show that to rapidly solve this grouping problem, the human visual system relies on geometric properties of the bounding contours such as symmetry and closure and not on the texture of the two-dimensional regions they partition. The independence of texture and contour in rapid perceptual grouping suggests that the visual system codes and links contours into coherent shapes before surface properties are conjoined.