Correcting a Digital Map Data Base with Scene Analysis Concepts for Automated Screening

摘要

To date, research in automated image classification has emphasized the application of statistical pattern recognition to multispectral image data. After some success in specific problem areas, progress has been limited due to many sources of spectral variability which have been difficult to model. Apparently, more robust approaches are required if computer image analysis techniques are to successfully augment the manual image interpretation capabilities exhibited by expert human image analysts. The long-range goal of our research is to define concepts which will lead to more successful strategies for automated scene analysis. We seek to do this in a fashion which will support incremental progress - tackling simple problems first in a manner which will permit more difficult problems to be addressed later. Contents: Design of a Research Strategy for Automated Screening; Map-Guided Scene Analysis Software Modules; Automated Screening Experiments; Research Data Set.