Integrating spectral and textural information for land cover mapping

Experiments aiming at integration of spectral and textural information showed great potential in image classification. The objective of this study was to assess applicability of variogram-derived texture measures in integrated optical and microwave images classification. The study was conducted using Landsat TM and ERS SAR images, covering Kolno Upland, an agricultural region in northeastern Poland. A set of textural parameters was derived within a moving window of an adapting size (determined by range of a variogram). Decision binary trees (DBT) were used to assess the potential of texture derived from variograms, cross variograms and pseudo-cross variograms to discriminate between land cover classes. Accuracy of DBT constructed on both spectral and textural parameters increased by as much as 10%, comparing with the accuracy of standard per-pixel classification. However, as the maximum likelihood decision rule did not perform well, further methods for improvement of the variogram-derived texture classification are discussed.