Image Filtering by Kriging Analysis

Abstract

Digital images contain so much information that it is difficult to identify and interpret spatial structures that they might contain. We have investigated kriging analysis to reveal structure in imagery and to estimate its spatial scales separately. We applied the analysis to part of a SPOT image of a dissected landscape in the south eastern United States. Variograms for the three wavebands in the imagery and the normalized difference vegetation index (NDVI) showed variation on two distinct spatial scales, which we modelled with a nested (double) exponential function, with effective ranges of 140 to 300 m and 2340 to 3400 m. By kriging analysis we estimated the separate contributions of the short- and long-range spatial components. The maps of the short-range components seem to represent a patchiness in the ground cover, whereas the long-range component seems to Refslect the coarser pattern imposed by the gross physiography. The paper is illustrated with the results of analysing the NDVI.KeywordsNormalize Difference Vegetation IndexOrdinary KrigingImage FilterKriged EstimateNugget VarianceThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.