Seismic pattern recognition via predictive signal/noise separation

Manual stratigraphic interpretation of modern 3-D seismic images is a tedious and timeconsuming process. We present a method based on nonstationary predictive signal/noise separation for automatically recognizing the occurence of an arbitrary, predefined pattern, or facies template, in seismic images. Similarity of local data windows to the facies template is measured by an attribute which has an easily interpretable physical meaning. The method is tested on 2-D synthetic and real seismic images, and is shown to reliably detect the presence of unconformities in both. An extension of the method to 3-D should be quite straightforward, and early performance assessments hint that the extension will not be hindered severely by computational issues.