Numeric and symbolic data combination for satellite image interpretation

Describes an algorithm allowing cooperation of different approaches and various data sources (image and out-image data) for classification purposes. The authors take into account expert knowledge about the context of each searched class, this context being described using out-image data (Digital Elevation Model, soil map, roads...eventually a Geographical Information System). The expert knowledge is expressed using terms showing uncertainty or frequency value about contexts for the classes. This knowledge is used to produce for each point potentiality degrees for all searched classes. This set of potentiality degrees is considered as one information source (based on expert knowledge and related out-image data). The authors then consider that they have two information sources about the searched classes (potentiality degrees and belonging degrees) to reach the final decision. The combination process uses a Mycinlike certainty factors method or Dempster-Shafer orthogonal rule for combination of uncertain information sources. This approach is used to resolve a particular problem in the frame of a geographical information system and to produce a potentiality map.<<ETX>>