Using a Context-Inclusive Approach to Process Statistical Queries in Raster Data: A summary of results

Many statistical queries such as maximum likelihood estimation involve finding the best candidate given a set of candidates and a quality estimation function. This problem is common in important applications like landuse classfication from remote sensing raster data. Such problems can be computationally challenging due to the significant computation cost to evaluate the quality estimation function. A recently proposed method of multiscale, multigranular classification evaluates the quality estimation function for various candidates independently before comparing candidates and hence suffers high computation overhead. In contrast, we propose a context-aware approach that selects a candidate based on a context and reduces computation cost though with some loss in accuracy. Experimental results are provided to show how the approach works.