Seeded growing algorithms for salt body segmentation in post-stack seismic data
MetadataShow full item record
- Studentoppgaver (TN-IDE) 
For a more accurate geological model, it is crucial to have proper segmentation and delineation of structures. Salt structures are known to be difficult to segment given their chaotic nature; however, there are methods that isolate the salt borders well. Delineation of salt structures in post stack seismic data is an inherently di cult problem and is of great value when detecting potential reservoirs. This is the case in regions such as the Gulf of Mexico. There exist several approaches where an image ltering algorithm is used to detect such structures; such is the case with well known seismic attributes like variance, coherence, amplitude contrast, etc. These attributes usually are a good indication of the anks of the salt structure but struggle in segmenting the salt body as a whole. In our work, we propose an automated seeded growing method that will segment the salt bodies from the other structures in seismic data. We Used the edge volumes as input to get a good indication of the anks and borders of our salt structures. We build our method upon the conventional seeded growing approach, and expand it with hybrid smoothing methods such as median, mean, Gaussian, adaptive median. Our alternating criteria, in this case, is the amount of chaos detected. Finally we detected the discontinuous boundary by evaluating the growing directions of our seed points and terminate based on unbalanced behavior. We implemented our method using Matlab and tested it using dataset from the Gulf of Mexico. By running our algorithm, we got an automated detection of the seeding within the salt bodies, clear segmentation of the salt bodies and a termination at the anks. The results show that even given the noisy nature of the salt images, the method is able to segment the salt bodies entirely and consistently. We managed to terminate only at the boundaries of the salt bodies and not before or after. This gives a more consistent segmentation, which is evident even in the case where the structures boundaries are disconnected. Our approach of estimating the boundaries when not present and looking at unbalanced growing, results in a consistent segmentation nonetheless. Evaluating all the stages in our algorithm, we nd that the computation complexity is bounded by O(NlogN) The approach of a seeded growing algorithm to segment salt bodies proves to be useful, and given a proper input volume is able to isolate the salt body as a whole. The approach presented here, with the modi cations, is e ective at this task only when combined with methods of noise removal/reduction and boundary termination estimation for existing and discontinuous boundaries. The hybrid smoothing approach has also proven to be a very useful combination with the growing method as it alternates smoothing techniques to optimize the segmentation.
Master's thesis in Computer science