Texture features for classification with ERS/JERS composites

Using the nearly-global coverage of the ERS and JERS SAR satellites regional and global classification algorithms are now possible. Because they provide information at 2 distinct wavelengths, their combination has been useful for previous efforts towards such global classification algorithms. This paper extends that work with the use of various textural measures that allow the classification procedure to discriminate more classes (up to 11 so far) and with high accuracies (>85%). The texture features used were: grey-level c co-occurrence matrix measures, and a measure due to Ulaby that calculates a speckle-corrected standard deviation. Remarkably, the classification accuracies and number of classes is comparable to those obtained when using fully-polarimetric SAR data