A comparative analysis of spatial indices and wavelet-based classification

Abstract

Spatial indices measure the geometric arrangement of land use and land cover classes at various scales and are computationally adaptive with wavelet transform coefficients. Decision rules built on permutations of three spatial indices – energy, log energy and Shannon's diversity – are used to improve the accuracy of multi-resolution hierarchical wavelet-based classifications. Comparisons are made with classification results derived from other texture measures, as well as with classification results calculated from more conventional per-pixel techniques. The first decision rule that computes one distance for each class across all bands to select the shortest distance was the most effective decision rule, resulting in the highest overall classification accuracy using the energy spatial index.