Perceptual Structure and Shape from Texture
Abstract
There were two major objectives of our research under grant AFOSR-88 -0219. The first of our research concerned perceptual grouping. The goal here is to segment an image into its perceptual components or segments. Such perceptual structure may exist at a range of resolutions. In the second part of our research we were concerned with a computational theory for an integrated representation of texture that takes into account many relevant aspects of texture, rather than just the properties of the texture elements at one level of resolution addressed in the past work. We have made progress towards both of the above mentioned objectives. We have extended our integration approach for perceptual grouping to extract perceptual structure in gray level images. Analogous to the original approach which was applied to dot patterns, the extended approach infers the structure by integrating evidence from region boundaries and region interiors. A large region considered homogeneous may contain regions deemed homogeneous at their own scales. The evidence for region interior is derived by using the multiscale region detector we developed earlier for the shape-from-texture problem. Region borders are also extracted at multiple resolutions, using a nonisotropic edge operator. In integrating the region interior and border information, the region boundary is forced to be smooth using explicit constraints to that effect.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 14, 1989
- Accession Number
- ADA207873
Entities
People
- Narendra Ahuja
Organizations
- University of Illinois Urbana–Champaign