The volumes of modern gridded data collected by the geophysical exploration industry are often large and can benefit from methods for image decomposition, pattern analysis, and interpretation. Automated pattern-recognition methods can be useful for both seismic and potential-field images. In both of these cases, it is important to extract quantitative attributes useful for further data analysis and inversion for the subsurface structure. In this paper, we investigate one of such pattern-analysis approaches called “skeletonization”.
Although using different physical fields and models, many types of 2-D geophysical images, such as seismic sections and slices, gravity and magnetic maps, possess a number of similar geometrical features. These features can be expressed by linear continuity, branching, amplitudes, widths, polarities, curvatures, orientations and/or other attributes and can be subdivided into “background trends” on top of which some kinds of “anomalies” or “wavelets” can be recognized. The types of spatial dimensions of the images may also vary, ranging from the usual distances, elevations and depths to travel times and travel-time lags. Automatic identification of such spatially-connected wavelets and measurement of their parameters is the general objective of skeletonization.
This paper describes a new, azimuthally-uniform skeletonization approach to 2-D potential field data. The majority of the material in the following sections was originally published as Paper A-3 in Summary of Investigations 2012, Volume 1, Saskatchewan Geological Survey Miscellaneous Report 2012-4.1 (Gao and Morozov, 2012). This material is reproduced here, with modifications, with permission from the Saskatchewan Geological Survey.