A comparison of classification algorithms in terms of speed and accuracy after the application of a post-classification modal filter

Abstract Four different supervised classification schemes—minimum distance, decision tree, maximum likelihood and a modified minimum distance classifier (the ‘deviant distance’ algorithm)—were applied to Landsat Thematic Mapper imagery. They were compared in terms of speed of computation and classification accuracy. The processing time required by each classifier was noted and accuracy of each calculated from contingency tables. Modal filters (3×3 and 9×9) were then applied to the classified images and the processing times and classification accuracies were compared. In this empirical study it was found that although the maximum likelihood algorithm provided the most accurate classification, the use of a faster algorithm, such as minimum distance followed by the application of a modal filter, could provide classifications of similar accuracy in less than half the time taken by the supervised maximum likelihood algorithm.