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：2016
卷号：XL-3/W4
出版社：Copernicus Publications
摘要：Object-oriented or segment-based classification approaches for remote sensing applications integrate not only the spectral signature but also shape and topological characteristics of segments. To use their shape related features, segments have to be as close as possible to the desirable object shape. The segment boundary's quality is decisive for the classification output - in order to delineate two classes and to avoid spectrally mixed segments. Despite its importance, no standard has yet been accomplished to address the segmentation's quality evaluation. This contribution presents further investigations on this topic based on boundary accuracy by using a distance dependent weighted quality rate and integrating other quantities yielding a combined quality measure. Furthermore it analyses the dependency on reference objects' quality. The reviewed quality rates are not only used for evaluating the segmentation, but are also used for the decision which segmentation level within a multi-scale segmentation should be used for classification of a certain class
