A Clustering-based Approach to Map 3D Seismic Horizons

Abstract

We describe a clustering based methodology to map 3D horizons automatically. From a Cosine of Instantaneous Phase version of the entry 3D Seismic Data, we represent the volume voxels by feature vectors that are windows of vertical neighboring voxels. We vary the window sizes, creating many representations for each voxel, and creating many datasets of feature vectors, organizing the vertical windows according to its size. From the datasets we create many clustering procedures creating distinct sets of clusters, so that voxels are represented by the clusters where its corresponding vertical windows were classified. Based on these clusters, we compute a similarity function that is naturally non-local and auto-adaptable, optimized for each particular seismic data. This similarity function is the measurement that decides if two arbitrary voxels compose the same horizon. The experimental results indicate the efficiency of the proposed method and illustrate its advantages.