Resource information
The paper investigates the performance and potential of object based image analysis technique (OBIA) for land cover information extraction from high resolution satellite datasets. Efficiency of the technique has been assessed in different urban land cover situations using merged CARTOSAT-1 and IRS-P6 LISS-IV image subsets of Dehradun, India. Multi-scale iterative Nearest Neighbour classification has been applied. Classification results have been quantitatively evaluated. Results reveal that OBIA is a fast, simple, flexible and efficient semi automated tool to deal with high spatial and spectral heterogeneity inherent in high resolution urban datasets. Integration of shape and texture characteristics with traditional spectral signatures, and use of multiple object levels in the classification process yielded classified map units having high correspondence with real world geographical entities. High accuracies were achieved for spectrally and spatially distinct classes like built up structures and tree covered land. Performance of the technique is impaired in case of classes having overlapping spectral characteristics. The parameters for segmentation and class descriptions developed for one area were successfully transferred to the other with minor adaptations. Utility of the technique is further enhanced by flexibility of visualisation of maps at different levels of classification hierarchy and immediate integration of the classified product in GIS environment with direct realization of vector maps for spatial analysis.
