Multiple knowledge sources and evidential reasoning for shape recognition
- 30 December 2002
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 624-631
- https://doi.org/10.1109/iccv.1993.378153
Abstract
A shape recognition approach is presented. Uncertainty handling, combining, and propagation form the heart of the method. Multiple knowledge sources extract information from the segmented image and increase knowledge about undefined shapes. Knowledge sources have to be tuned to discriminate shape classes, and a critical number of independent knowledge sources guarantees the classification. Information provided by the knowledge sources is stored in the Shafer form of probability mass assignment. Dempster's rule is used to update belief in classes. A brief theoretical overview is given. Combined with a heuristic, this method achieves interesting results as well as a short execution time. An example derived from an application in the PROMETHEUS project, consisting of traffic sign recognition on a motorway, illustrates this methodKeywords
This publication has 6 references indexed in Scilit:
- Artificial intelligence dialects of the Bayesian belief revision languageIEEE Transactions on Systems, Man, and Cybernetics, 1989
- Smart sensing within a pyramid vision machineProceedings of the IEEE, 1988
- Parallel algorithm for corner finding on digital curvesPattern Recognition Letters, 1988
- Recognition of Hierarchically encoded images by technical and biological systemsBiological Cybernetics, 1987
- Curve Similarity via SignaturesPublished by Elsevier ,1985
- Hierarchical Contour Coding And Generalization Of ShapePublished by SPIE-Intl Soc Optical Eng ,1984