Feature Based Grouping to Detect Suburbia

Abstract

In this and the following chapter, we focus on detecting suburban regions among others. Although it is part of land use classification problem, we had to introduce specific methods to detect these regions in a robust manner. The direct three class classification (urban, rural, residential) approach was less successful in this case, largely because suburban regions bridge the other two in our feature space much as they do on the ground. Therefore, in an attempt to extract suburban regions, we introduced an enhancement based on the principles of perceptual organization. Perceptual organization is that process, or a set of processes, by which a vision system (natural or artificial) organizes detected features in images based on various Gestaltic clues. Perceptual organization is therefore the ability to impose structural regularity on sensory data, grouping sensory primitives having a common underlying cause. We introduced a spatial coherence constraint and performed grouping in the feature space. Via this novel perceptual grouping approach, the results improved significantly. Hence, besides the structural approach to land classification, our new spatial coherence method based on perceptual organization principles also offers very promising results by combining the feature and image spaces.