Estimating texture independently of tone in simulated images of forest canopies

Description/Abstract

Tone and texture are two fundamental characteristics of remotely sensed images. Current research on the remote sensing of tropical forest biomass uses the tone (i.e., backscatter) of Synthetic Aperture Radar (SAR) images as this is related directly to biomass (albeit up to the backscatter/biomass asymptote). As a tropical forest canopy ages so its unevenness increases, progressing from smooth to rough. Therefore a measure of SAR texture that is independent of SAR tone has the potential of increasing the biomass maxima that can be estimated with SAR data. This experiment used simulated SAR images designed to reproduce forest canopies and different patterns of tone (or contrast) and texture (or clumpiness). Twenty six texture measures (derived from local statistics, the grey-level co-occurrence matrix (GLCM) and variograms) were calculated for these simulated images. Measures sensitive to texture (clumpiness) and/or tone (contrast) were identified using Analysis of Variance (ANOVA). Seven texture measures were recommended for the estimation of tropical forest biomass with SAR images.