Texture based information extraction from high resolution images using object based classification approach

High resolution satellite images have been actively utilized for information extraction. Object oriented classification approaches based on the segmentation are being adopted for extraction of variety of thematic information from high resolution satellite images. Object oriented classification method is composed of two successive processes. Firstly the image is subdivided into different objects based on the spectral and spatial heterogeneity in segmentation process. Then objects are assigned to a specific class based on the detailed description of the class in classification process. This paper describes the homogeneity parameters including scale factor used for segmentation and utility of texture information for object based classification. Various GLCM texture features are extracted from the segmented image and these features are further used in classification process. The cartosat-1 satellite data has been segmented and classified into six land use/cover classes using eCognition software. The satellite image has been segmented at various scales parameter out of which scale 50 has been found to better which produces the overall accuracy of classification 85.16% and kappa coefficient 0.8115.