A new method for more efficient seismic image segmentation

Abstract:

Seismic images of the subsurface are often very large and tedious to interpret manually; as such, automatic segmentation algorithms can be highly useful for tasks such as locating large, irregularly-shaped salt bodies within the images. However, seismic images present unique challenges for image segmentation algorithms. Here, a new segmentation algorithm using a ``pairwise region comparison'' strategy is implemented and tested on seismic images. Numerous modifications to the original algorithm are necessary to make it appropriate for use with seismic data, including: (1) changes to the nature of the input data, (2) the way in which the graph is constructed, and (3) the formula for calculating edge weights. Initial results, including a preliminary 3D implementation, indicate that the new method compares very favorably with an existing implementation of the eigenvector-based normalized cuts approach, both in terms of accuracy and efficiency.