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：2007
卷号：XXXVI-3/W49A
页码：117-122
出版社：Copernicus Publications
摘要：Detection and quantification of regularity patterns are important structural aspects for object-oriented classification of images for geo-databases updating. Four image processing methods are analysed and evaluated for this purpose: semivariogram analysis, the Hough transform, the histogram of minimum distances, and Fourier space descriptors. In addition, several features are extracted from each method and evaluated for classification of regular and non regular parcels in a rural environment. The classification has been performed by using the C5 algorithm, based on data mining techniques. A total of 276 objects have been evaluated using the cross- validation method. After selecting the most discriminant features, a land use object-oriented classification has been performed, which includes some spectral and textural features in the model. The results show that the features based on the semivariogram and the Hough transform are the most efficient for detecting regularity patterns. The combination of the three groups of features (spectral, textural and structural) clearly improves the classification of parcels, which is encouraging for automated land use cartography updating
