Impervious surface detection from multispectral images using surf

Abstract

Detection of different regions like impervious surfaces, vegetation and water from a multispectral satellite image is a complex task. This paper introduces a novel idea for impervious surface detection from multispectral images using SURF descriptors. To determine the efficiency of the proposed system, a comparative evaluation is done with other two techniques, namely histogram based and spectral-value-based technique. The result shows that the proposed system outperforms the other two techniques in detecting impervious surfaces like buildings and vehicles with an accuracy of 80.48%. The histogram-based technique and spectral-value-based clustering obtained an accuracy of 61.89% and 68.29% respectively. However, in classifying vegetation the other two techniques outperforms SURF descriptors. The histogram based technique gives an accuracy of 86.46% and an accuracy of 94.35% is obtained by using the spectral-value-based clustering. Whereas SURF based technique gives only an accuracy of 50.71%. © 2014 Springer International Publishing Switzerland.