Comparison of techniques for measuring cloud texture in remotely sensed satellite meteorological image data

Abstract

This investigation has attempted to discover appropriate texture descriptors and to reveal more clearly the importance of texture analysis techniques for multispectral cloud classification. The textural features considered in this study include both spatial and frequency features. The spatial features were mainly those based on spatial grey-level difference statistics and circular Moran autocorrelation measures. The frequency features were those based on summed energies of polar co-ordinate Fourier power spectra and entropy-based measures of the spatial distribution of frequency entries in the polar spectra. Some other textural features, such as the Roberts gradient measure, were also investigated. The work was performed with TIROS-N AVHRR image data acquired in the late spring of 1979 over areas near the British Isles. The results of the evaluation and corresponding conclusions show which individual, and group of textural features, appear the most appropriate for aiding multispectral cloud classification.