Automatic boundary detection using potential-field data

The boundary detection problem uses image transformations and gradient properties to identify the location of possible subsurface boundaries from geophysical data. Automatic methods exist that show the boundary locations either as pixels in a raster image or represent those lineaments as vectors. We present a framework to extract lineaments as vectors while maintaining a space filling mesh. We formulate the solution as one of an optimizaton problem with two separate steps. A lattice of nodes is first distributed horizontally on an image of the data so that the generated mesh aligns with features in the data. To deal with the noise in the data, we require the second step of vertical node optimization. This is achieved by embedding the 2-dimensional lattice and data into a 3-dimensional domain. The resultant optimized node heights allow for a more reliable extraction of mesh lines concordant with lineaments. We demonstrate the framework with a synthetic example and two airborne magnetic data sets.