Texture-based techniques for interpretation of seismic images

Seismic techniques are among the most popular and successful methods for imaging the Earth's subsurface. Consequently, these techniques play an important role in the exploration for hydrocarbon deposits within the Earth. Seismic data are typically displayed as a two-dimensional image of stacked traces placed side-by-side. The vertical direction of such a display is time, which can be converted to depth once the signal velocities are known. The horizontal direction is linear distance on the surface of the Earth. The interpretation of a seismic image involves, among other things, the recognition of certain geologic patterns such as faults, salt domes, strong reflectors, etc. Some of this can be achieved by identifying large zones of common signal texture. Such zones, are often associated with major geologic events related to the depositional environment of the constituent sediments. As such, their identification is an important part of the overall interpretation task. In this paper, the authors describe three knowledge-based segmentation techniques and compare their performance on a real offshore seismic section from the Gulf of Mexico. The first technique is based on a run-length statistics algorithm extended by a decision process, which incorporates heuristic rules to influence the interpretation. The second and third techniques are based on texture energy measures algorithms augmented by two knowledge-based classification processes. The knowledge-based process of the second technique is controlled by a parallel region growing algorithm and that of the third technique is controlled by an iterative quadtree splitting algorithm.