Geological bodies of various kinds may be found in the underground. Automatic retrieval of such objects in seismic images would be a rapid, objective help for interpreters. As shown in our application on real data, our methodology allows to retrieve heterogeneous objects. Texture characterization and a two-stage segmentation first produce object-presence probability images; then, object filtering according to predefined criteria is applied to postprocess the images, yielding heterogeneous geobodies extension and characterization.

This is poster number 48 in Poster Session

Authors:

Pauline Le Bouteiller (Université Pierre et Marie Curie)

Jean Charléty (IFP Energies Nouvelles)

Florence Delprat-Jannaud (IFP Energies Nouvelles)

Christian Gorini (Université Pierre et Marie Curie)

Didier Granjeon (IFP Energies Nouvelles)

Keywords:

image segmentation, machine learning