Object-Based Analysis

In the preceding chapter, the techniques available to exploit the spectral content of our remotely sensed imagery were discussed. The methods and issues reviewed in Chap. 10 did much to demonstrate the unique and useful sources of information that can be gained from the expanded spectral domain offered by hyperspectral sensors. Although essential, spectral data alone does not provide a complete picture of the environmental system to fully understand process or guide environmental decision making. One under-utilized quality of an image that can add an additional dimension to the study of the environment is the explicit spatial arrangements, juxtapositions and patterns exhibited by our spectral measurements. These unique arrangements place spectral data in a geographic context where spatial descriptors such as shape, perimeter, and texture together with other geographic variables define objects in the image that add an element of knowledge into the classification problem. These image objects are derived exclusively from the spatial relationships found in the image and with this additional knowledge they present criteria beyond the spectral signature which enriches the image classification process. Object-based analysis offers new possibilities that may extend the role of remotely sensed data in complex mapping and assessment applications. In this chapter, we will examine the object-based paradigm and review the fundamental aspects of image classification based on the analysis of image objects.