Abstract

With the advent of higher-resolution remote sensing imagery the need for more efficient multi-scale image analysis methods has grown more than ever. This paper presents a superpixel- based methodology for hierarchical image analysis. The use of superpixels as pre-processing block reduces the execution time of the analysis and also provides a good starting point to build meaningful objects at different scales through a merging process. At each scale a set of most representative features are extracted, and then regions of interest are identified through a classification method. Only classes of interest are analyzed in a later segmentation step.