AUTOMATIC EXTRACTION OF RIVERS IN SATELLITE IMAGES USING GEOMETRIC ACTIVE CONTOURS

This work aims to define and test a method for the extraction of rivers in orbital images for regions that are seasonally flooded, ie, large areas containing more water bodies besides the river, such as Pantanal. In the proposed methodology, several tools from the area of Image Analysis and Computer Vision have been employed, performing a preprocessing, followed by a topological modeling that is built upon a skeletonization process followed by an analysis of this skeleton. Lastly, the methodology selects structure that represents the river, by performing a topological filtering. This process is responsible for the selection of points that initiate the process of delineation of rivers using a geometric active contour method, called .Level Set Method.. The methodology was evaluated qualitatively (visual) and quantitatively (numerical) using the criteria of completeness and correctness in a series of real images of the Pantanal region. The edges extracted from rivers, were projected onto the original images, thus allowing a qualitative assessment. With respect to the numerical results for the criteria of completeness and correctness, these were always above 80%, which shows that the methodology is very effective and robust for the community that needs to perform feature extraction in remote sensing image