Object-oriented Image Analysis for the Identification of Geologic Lineaments

In this work, a methodological framework employing image processing techniques has been developed and applied on a LANDSATETM+ image of Alevrada, Greece. Pre-processing of the satellite image was followed by edge detection performed on band 5 of the ETM+ image in order to obtain proper input data to the lineament identification system. NDVI, PCA and ISODATA unsupervised classification were applied for the discrimination of the land cover classes. The input data layers to the lineament identification system included (1) an edge map from the EDISON algorithm, containing the edges to be classified, (2) the geologic layers derived from the pre-processing phase, (3) the initial ETM+ image and derived thematic products. Image segmentation was based on the multi-scale hierarchical segmentation algorithm for the extraction of primitive objects to be classified during the classification process. Six segmentation levels were created and their parameters were selected though a trial-and-error procedure. The design of the knowledgebase involved in the definition of classes / sub-classes of the scene, by spectral and geometric attributes, texture, spatial context and association on each segmentation level. Fuzzy membership functions and the Nearest Neighbor Classification were used for the assignment of primitive objects into the desired classes combining all participating levels of class hierarchy. The output result of the system was a classified lineament map containing the inherent geologic lineaments of the study area (faults) and the lineaments, which were not identified as faults (non-interest lineaments). From the classification stability map, it was inferred that there was a high degree of coincidence of the extracted results of the knowledge-based scheme and the tectonic map of the area.