Evaluation of image comparison methods for complex textures

Abstract

In the O&G industry, interpreters may be assisted by image retrieval techniques when searching for relevant geological features in a seismic dataset, such as image patterns and structures. This can greatly accelerate the interpretation process by enabling the interpreter to relate the observed features with potential analogs from other interpretation projects. In this paper we evaluate the performance of different image comparison methods for seismic image retrieval. Most methodologies proposed in the literature assume that query and database images have comparable sizes. Our analysis includes methods that are able to compare images with arbitrarily different sizes which not only makes the retrieval application more general but also allows to answer queries not addressed by usual methods such as within, contains and resembles. The results in a public seismic dataset show a gain of up to 180% for P@1 in comparison to traditional methods.