An investigation of textural characteristics associated with spectral information for land use classification

Abstract

The objective of this study was to investigate improvement of classification accuracy using synergism between textural features and spectral information. Satellite data used in this study are multispectral SPOT HRV, Landsat-TM, and JERS-1 SAR images. Spectral information applied for data compression, is standard principal component analysis, while speckle noise present at JERS-1 SAR image was reduced using wavelet transform. The first order statistic of variance and the second order texture statistic of entropy found in the literature were used. Several datasets were generated using spectral extraction, textural features, and their combination. Based on the maximum likelihood classifier, land use categories of the study area were discriminated. The result shows that combined use of spectral and texture information together significantly improved the accuracy of land use classification.