An Object-Oriented Pattern Recognition Approach for Urban Classification

In contrast to subpixel and per-pixel image classification approaches, object-based image analysis (OBIA) attempts to exploit spatially and spectrally similar groups of pixels in order to identify objects within an imaged scene. Object-based approaches are most applicable to high spatial resolution data, where objects of interest are larger than the ground resolution element. Such objects in urban scenes could be related to natural features of urban landscapes (e.g., trees and lakes) or man-made features (e.g., buildings or roads). This chapter introduces readers to the principles of OBIA and demonstrates how it can be applied to achieve satisfactory accuracy in urban mapping. We employed two case studies with two example subsets extracted from Quickbird multispectral satellite data and demonstrated two object-based analysis procedures, namely decision rule (i.e., membership function) and nearest neighbor classifiers. The object-oriented classification employed to identify urban classes in this chapter is specifically based on Definiens software and the various routines it contains to achieve OBIA. However, an overview of the object-oriented approach, parameters generally available for object analysis, image segmentation procedure, rule set approaches, nearest neighbor classifier using training samples, and limitations/uncertainties associated with object-based techniques reported in this chapter are applicable to urban mapping using any OBIA software.