Segmentation in Acronym,
Abstract
The role of segmentation in the ACRONYM vision system is to detect, group, and represent low-level image structures such as edges and curves without detailed knowledge of objects in the scene or of viewpoint. These structures are then passed to higher levels of processing which do use such knowledge to predict the appearance of objects in images and then to match the predictions with the observed structures. Currently this matching takes place at the level of ribbons, the perspective projection of the generalized cylinders from which ACRONYM's models are built. The three steps in finding ribbons are the detection and linking of edges, the segmentation of linked edges into curves, and the grouping of curves into ribbons; of these, the first two have recently been significantly improved, and revision of ribbon grouping operations is under way. Formerly, linked edges were segmented only into straight lines, so that providing curved data for grouping into ribbons should lead to superior performance. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1982
- Accession Number
- ADP000128
Entities
People
- David H. Marimont
Organizations
- Stanford University