Efficient Geospatial Analysis of Remotely Sensed Images by Means of Linear Feature Extraction and Combination

An efficient and valid interpretation of very high resolution images, either SAR or optical, must take into account the spatial details of the scene. As a result, there are many options to the very simple pixel-by-pixel classification scheme which applies to coarse resolution remote sensing data sets. One option is the exploitation of spatial features, like for example textures, that connect each pixel with its neighborhood (Dekker 2003). Another option is multi-scale analysis, where scales, and thus contexts with different size, are jointly considered to capture details at various levels (Benediktsson et al. 2003). A third option, explored in this chapter, is to extract significant yet simple geometrical features and use them (possibly in combination with spectral features) to improve the understanding of the scene (Xin et al. 2007). It is a separate approach, other than standard classification, but it can be used also for classification and change detection (which is essentially multitemporal classification).