Performance level of accuracy measuring methods in classified remote sensing images as applied to the built environment.

Abstract

Assessing the accuracy of a classified image is an essential task that gives the user apriori information of the overall reliability of subsequent analysis performed with such classification methods. This research seeks to carry out an assessment of the accuracy measure for evaluating the integrity of the result of image classification using the overall accuracy and the Confusion Matrix. The effect of the size of the defined training site on the accuracy of the resulting classified image has also been examined. LandSAT image of part of South Western Nigeria was used in this study with three different classification methods (Maximum Likelihood, Mahalanobis distance and minimum distance Classifiers). The results obtained shows that the use of the confusion matrix gives a better analysis of the level of reliability of the classification than the use of chance adjusted indices or overall accuracy.