Abstract

This study aimed at an analysis and validation of image classification techniques in mapping urban areas using high resolution satellite imagery. It’s objectives were to identify and extract regions of interest (ROI) from the study area subset of the high-resolution imagery, to perform image classification using Maximum Likelihood Classifier, Support Vector Machine, rule-based, example-based and index-based classifiers and to evaluate the performances of Maximum Likelihood Classifier, Support Vector Machine, rule-based, example-based and index-based classifiers using error matrix, kappa, correlation coefficient, standard deviation, standard error, mean square error and root mean square error. The methodology covered data acquisition of high-resolution satellite image data, data preprocessing for the acquired image data, image classification and image classification assessment with error matrix, kappa, correlation coefficient, standard deviation, standard error, mean square error and root mean square error. The classification results obtained from maximum likelihood, support vector machine, rule-based, example-based and index-based classification indicated that maximum likelihood and support vector machine classifiers achieved higher classification values for agricultural area and commercial area, while achieving lower values for the classification of open space and residential areas. Rule-based, example-based and index-based classifies, all had the values for agricultural, commercial areas, open space, industrial and residential areas in similar range. In the classification of waterbody, all classifiers had all their values in the same range. Using the final ranking of the results from error matrix, kappa, correlation coefficient, standard deviation, standard error, mean square error and root mean square error, example-based classification ranked as the best in the group, rule-based classification ranked second best, support vector machine classification and index-based classification ranked third best, while maximum likelihood classification ranked fourth. The study recommends the example based object-oriented classification approach as it is a robust and efficient tool for mapping different features within the settings of an urban landscape.