The Use of Models in Image Analysis.
Abstract
Several of the automatic components of an interactive imagery screening and target detection system are studied. Components include a module for detection of primitive image features without model hypothesis, a module for generation of hypotheses given that a particular model for the data has been evoked by primitive feature content, and a module for testing the model generated hypotheses against the data. Only shape features extracted from boundary segments are considered in this work. Features used are straight, circular, and parabolic edge segments and points of intersection or points of high curvature on edge segments. Feature extraction algorithms are discussed. Two different model types are considered--problem reduction representations (PRR) which correspond to grammar models and iconic models which correspond to cartographic data bases. Once primitive features from an image are aligned with a potential model the model generates hypotheses about the data which can be either verified or denied. Testing of hypotheses is done via curve fitting or template matching. A new approach to registration is presented and documented via experimental results. The registration approach enables the use of cartographic data bases as models for use in image analysis and object detection. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1979
- Accession Number
- ADA067166
Entities
People
- George C. Stockman
- Steven H. Kopstein