An assessment of some small window-based spatial features for land-cover classification

Fourteen small (3/spl times/3) window-based spatial measures are applied to a SPOT HRV multispectral Band 3 image to extract spatial features for the rural-urban fringe of Metropolitan Toronto, Canada. The spatial features are combined with the original image to identify 12 land-cover classes. Four classifiers were used in the study. Results show that when nine of the spatial features are in turn combined with the three original images, significantly improved classification accuracies (at the 95 percent confidence level) are obtained compared with using only the multispectral information from the three original images.<
>