General Algorithm for the Semantic Decomposition of Geo-Images

The thesis presents an object oriented methodology for the semantic extraction of a geo-image which is defined by a set of natural language labels. The approach is composed of two main stages: analysis and synthesis. The analysis stage detects the main geographic components of a geo-image by means of the color quantification, geometry and topology of the geospatial objects. The result of this stage is a set of geoimages with intensities that are approximately uniform. The synthesis stage extracts the main geographic objects that have been identified and a labeling process in two levels (general and specialized), which is equivalent to consider both local and global information of a geo-image. The aim of the general labeling process is to associate a label of the adequate thematic to each region, taking into account the RGB characteristics of the image. In order to specialize each geographic object, we have proposed a specialization algorithm that considers geometric and topologic relations among them, represented in geographic application domain ontology. The obtained set of labels describes the geo-image semantics.