Image Analysis Algorithms for Semiautomatic Lineament Detection in Geological Outcrops

Digital Outcrop Models (DOMs) are increasingly applied to obtain large, statistically valuable datasets of 3D geological data. However, extracting the traces of geological features from outcrop images is still a time-consuming and subjective process, limiting the production of very large datasets and the reproducibility of the results. We are currently developing the DOMStudio workflow, a tool for speeding up and making more objective the detection of lineaments in outcrop images by combining image analysis with high-resolution image texturing on photogrammetric DOMs. In this chapter, we test and compare four algorithms for lineament detection not yet routinely used in geology-oriented applications: (1) a two-classes image segmentation based on Markov random fields, (2) the Difference of Gaussian filter, (3) the symmetry of image phases determined by log-Gabor wavelets, and (4) the congruency of image phases as obtained by convolving the images with complex shearlets. We introduce and discuss pre- and postprocessing workflows for extracting thinned, 1-pixel-thick lineaments useful to be converted in vector format. All the described algorithms leverage the image analysis capabilities of Matlab® libraries. We applied the algorithms to detect structural lineaments (faults, stylolites, veins, and fractures) in three images from DOMs collected in different geological settings: fractured sandstone, fractured limestone, and pseudotachylyte-bearing faults in granodiorite. We finally discuss the pros and cons of the four workflows in relation to the spectral signatures of lineaments against wall rocks in the examined outcrops.