In this research, we investigate the application of image processing and computer vision methods to geological images. The main objective is to develop a method for automatic horizons correlations across faults in 3d seismic data.
All decisions in hydrocarbon exploration and production are underpinned by subsurface models, which are obtained from structural interpretation of seismic images.
Artificially created sound waves are reflected between rock layers which have different velocities because of their variation in density and elastic properties. By recording reflection-time and amplitude of these waves, hypotheses about depth and properties of the boundaries between rock layers can be developed. Large differences between the velocities of layers yield to strong reflections being visible in the seismic images that are called horizons. Faults are fractures across which there is measurable displacement of rock layering. A fault is recognized by the discontinuities on horizon lines (see Fig. 1).
Manual interpretations of faults and horizon correlation are both costly and extremely subjective.
Automatic horizon trackers have been very good aid for structural interpretation of seismic data. But auto-trackers have problems of tracking horizons across discontinuities especially across faults due to insufficient information available in the seismic data.
Our research is aimed at correlating horizons across faults on the basis of reflection character and geological models. The task of correlating horizons across faults involves to gather seismic features on the two sides of the fault and then to find an optimal match between them using the low-level seismic features and geological model.
We investigate various feature computation tools and study geological knowledge required for designing the model. Furthermore, as matching problems have mostly exponential worst-case complexity, we explore various global-optimization methods.