A hybrid system for seismic section segmentation controlled by an iterative quadtree construction algorithm

Present a hybrid system for seismic section segmentation that combines texture image classification using artificial neural networks and rule-based seismic section interpretation. In this system, ideal texture events are first extracted from a digitized seismic section using two-dimensional (2D) convolution masks. These texture feature measures are the input later in segmentation to a set of trained artificial neural networks to produce quantitative image descriptions that can be used directly in rule-based seismic section interpretation. The overall segmentation process is controlled by an iterative quadtree construction algorithm under which the image data are analyzed and interpreted at different spatial resolution levels in an iterative fashion