A texture-based approach to the segmentation of seismic images

-A new method is presented for the texture analysis and segmentation of seismic images. The texture of a seismic image is described in terms either of seismic horizons' features (e.g. length, reflection strength, geometrical appearance), or in terms of Hilbert transform features (magnitude, phase, instantaneous frequency) or in terms of features related to the generalized runs. Seismic image segmentation rules are derived from examples by using minimum entropy rule learning techniques. Two new methods are presented for using geometric proximity to reference points in region growing. The first one is based on Voronoi tessellation and mathematical morphology. The second one is based on the so-called "radiation model" for region growing and image segmentation. Seismic image processing Hilbert transform features Minimum entropy rule learning Voronoi tessellation Horizon picking Run lengths Radiation model