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：2006
卷号：XXXVI-4/C42
出版社：Copernicus Publications
摘要：The automated interpretation of remotely sensed data is still performed largely on the basis of per-pixel image classification, i.e., the statistical analysis of each pixel's spectral value. The conventional per-pixel classification approaches, however, may have limitations to be considered when used with very high resolution imagery because these procedures ignore spectral reflectance characteristics of neighboring pixels. Object-based classification approaches can be an alternative to overcome these problems held in conventional per-pixel classification approaches. Although object-based classification techniques appear to cast a promising light on limitations of per-pixel approaches, further research is required to assess the quality of the classification results and the sensitivity of input parameters. The scale parameter controls the size of image objects and is the most important and critical issue because the quality of classification is directly affected by segmentation quality. This scale parameter should be a crucial factor to be considered in the use of object-based classification for alliance-level forest mapping. In this study, we adopted the graphs of local variance in an attempt to anticipate the optimal scale parameter for forest classification from multispectral Ikonos images. The local variance of image objects reached a threshold approximately at the scale parameter of 21 and the highest classification accuracies occurred around the threshold of scale parameter. The graphs of local variance may be used as a methodology to anticipate the optimal scale parameter before actual image classification
关键词：Object-based Classification; Optimal Scale Parameter; Graphs of Local Variance; Multispectral Ikonos Image
