EVALUATING POINT SET DISTANCE MEASURES FOR 2D SEISMIC ANALYSIS

Abstract

Recently, many works have investigated the use of texture features to support tasks such as seismic image retrieval, classification and clustering. These works aim to assist geoscientists in different applications by reducing the time, cost and potentially increasing the accuracy of the results. Our previous work indicated that for the texture-based retrieval of seismic images with different sizes, point set distance measures outperformed the classical rescaling approach which distorts the patterns and structures found in a seismic dataset. In this work, we consider a scenario in which we have many 2D seismic datasets and we want to run a quick in situ analysis to understand the diversity in the datasets and to cluster them according to their similarities. In our experiments, we investigate 6 point set distance measures on 12 seismic surveys comprising 304 seismic datasets. The results indicate that point set distance measures may be more suitable for the comparison of in situ datasets, showing a gain of 53% in comparison to the rescaling approach.