Empirical evaluation of distance measures for supervised classification of remotely sensed image with Modified Multivariate Local Binary Pattern

Texture classification is applied to remotely sensed imagery to get accurate results in terms of classification accuracy as every pixel is classified based on the collective relationship of the pixel with its neighbors. In this paper, Modified Multivariate Local Binary Pattern (MMLBP) texture model was taken up and supervised classification was performed on a remotely sensed image varying the distance measure used. A number of distance measures were taken up and applied to the marginal distribution comprising of one dimensional histogram called feature vector and the results were evaluated based on classification accuracy, inter cluster distance and intra cluster distance. It was shown that Bhattacharyya distance and Chi squared distances outperformed other distance measures.