Automatic segmentation of seismic data via knowledge-based image processing techniques

Three knowledge-based texture image segmentation systems are described, and their performance on a real test image is analyzed. The image is a small piece of a seismic section and the objective is to segment it into zones of common signal texture. The first system is based on a run length statistics algorithm extended by a decision process which incorporates heuristic rules to influence the segmentation. The second and third systems are based on texture energy measurement algorithms augmented by two different knowledge-based classification processes. The knowledge-based process of the second system is controlled by a parallel region growing scheme, and that of the third system is controlled by an iterative quadtree-splitting scheme. The results show that the third system, texture-energy measurement augmented by a knowledge-based quadtree-splitting process, provides a better segmentation of the test image than the other two