C. Venkatesan and R. Murugasan
Abstract
The urban environment has a variety of spectrally different materials, such as trees, concrete structures, roads, soil, grass, etc. High resolution satellite imagery is proving to be a cost-effective alternative to aerial photography and conventional surveying for the acquisition of information including urban base-maps consisting of land cover classes. Normally spectral information in the satellite imagery is used for the classification of urban features, but this is not found to be sufficient for classification of spectrally heterogeneous land-use classes, like built-up categories. In the present study, optimized textural parameters have been combined with spectral parameters as a hybrid methodology and improvement in feature extraction process has been investigated. The statistical technique of Grey level co-occurrence matrix (GLCM) was used in this study for providing information on texture variations. The utility of the textural analysis has been measured in comparison with normal multi-spectral per-pixel classification method. Using the next neighbors, co-occurrence matrix has been built. Haralicks second-order statistics has been applied to the co-occurrence matrix for extraction of features. It is found that there is an appreciable increase in the classification accuracy for urban environment when GLCM is part of the classification.