Traditional elements of image interpretation include characteristics of first order (tone/color) , second order (spatial arrangement: size, shape and pattern) and third order (height, shadow). In digital remote sensing third order image characteristics are considered a nuisance, while potentially useful spatial information has been usually ignored, due to lack of methodology and computational limitations. However few researchers have undertaken integration of spatial and spectral information for image classification: variogram-derived texture has been recently proved to increase the accuracy. The objective of this study was to assess the potential of variogram-derived texture measures applied to classification of high ground resolution imagery. The study was conducted using ADAR images acquired over the Santa Monica Mountains, dominated by chaparral vegetation. Textural parameters were derived from a moving geographic window (of a size determined by the range), as opposed to the commonly applied geometric window of a fixed size. Binary decision tree was used to assess the potential of texture derived from variograms, cross variograms and pseudo-cross variograms, to discriminate between land cover classes. Finally, images were classified using the most significant texture measures and the accuracy was compared with the accuracy of standard per-pixel classification. Overall classification accuracy increased by as much as 15%. Accuracy of homogeneous classes did not change, while significant increase was reported for highly textured classes. Further methods of improving accuracy using variogram-derived texture were discussed.
CITATION STYLE
Jakomulska, A., & Clarke, K. C. (2001). Variogram-Derived Measures of Textural Image Classification (pp. 345–355). https://doi.org/10.1007/978-94-010-0810-5_30
Mendeley helps you to discover research relevant for your work.