期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2002
卷号：XXXIV Part 2
页码：185-192
出版社：Copernicus Publications
摘要：In general, automatic object extraction systems are not to expect to deliver absolutely perfect results and, thus, for meeting predefined application requirements, a human operator has to inspect the automatically obtained results. In order to speed-up the time- and cost- intensive inspection, the system should provide the operator with confidence values characterizing its own performance. In this paper, an approach for self-diagnosis is presented which is part of an existing road extraction system. The design of the self-diagnosis tool is based on considerations from three different perspectives: From an operational point of view, one has to clarify which representation of the evaluation result is both convenient and effective for a human operator using the system. The choice of the evaluation criteria, on the other hand, is affected by the respective object model and extraction strategy on which the system is founded. Finally, an appropriate theory and methodology should be applied for evaluating and combining the criteria. We exemplify these aspects with the self-diagnosis tool of our road extraction system. Two different types of representation are incorporated to provide the user with the essential information about the extraction quality: A universal and a sectional representation. The criteria used for internal evaluation are derived from the object model. It is special attention payed to employ criteria not used during prior steps of extraction in order to reach a highly unbiased evaluation. Fuzzy-set theory is used as theoretical framework for knowledge representation for both extraction and evaluation. In order to analyze the quality of the self-diagnosis, we extracted roads in a test series of aerial images and matched the internally evaluated result, i.e., the labelled road sections, to a manually plotted reference. The comparison shows the benefits but also some remaining deficiencies of the self-diagnosis tool
