Automatic Extraction and Evaluation of Geological Linear Features from Digital Remote Sensing Data Using a Hough Transform

The Hough transform is an established tool for discovering linear features in images. The present investigation presents a new and specific algorithm for detecting geological lineaments in satellite images and scanned aerial photographs which incorporates the Hough transform, a new kind of a "directional detector," and a special counting mechanism for detecting peaks in the Hough plane. Three test sites representing different geological environments and remote sensing altitudes were selected. The first site represents sedimentary conditions of chalk beds on cherry picker photography; the second represents plutonic conditions of granite rocks on an aerial photograph; and the third represents tectonic fractures of carbonates, chalks, and cherts on digital satellite data. In all cases, automatic extraction and mapping of lineaments conformed well to interpretation of lineaments by human performance. Introduction Regional study of linear features such as faults, joints, folds, dikes, crustal fracturing, and lithological contacts, using aerial photographs and particular satellite images, has made important advances in geological research during the last few decades (Rowan and Lathram, 1980). Recognition of lineaments has been used for investigating active fault patterns in areas of difficult accessibility (Tibaldi and Ferrari, 1991), water resources investigations (Waters, 1990), mineral deposit exploration (Rowan and Lathram, 1980), and in the study of the structural or tectonic history of a region. The use of computerized, scanned data analysis for the detection of linear features from aerial photographs or digital satellite images can reduce to a minimum the bias of the subjective decision of the interpreter. Sijmons (1987) distinguished between two categories of computerized lineament processing. The first involves mainly enhancement of linear features using standard image processing methods [such as edge detection directional filters) for later visual interpretation while the second category involves automatic computer processing of the original digital data for the production of a lineament map. Obviously, the majority of studies which have been done up to date belong to the first category. For all of the cases in this category, interpretation of lineaments is still subjective, and considerable experience is required to detect weak features or to separate close ones. In order to overcome these limitations, an automatic objective procedure A. Karnieli, A. Meisels, and L. Fisher are with The Jacob Blaustein Institute for Desert Research, Sede Boker Campus, Ben-Gurion University of the Negev 84990, Israel. Y. Arkin is with the Geological Survey of Israel, 30 Malkhe Yisrael St., 95501 Jerusalem, Israel. is required. This type of procedure is not common (Oakes, 1987; Simon et al., 1989; Zlatopolsky, 1992). During the last few years, special interest has been shown in using the Hough transform algorithm for this purpose, but with limited success (Cross, 1988; Cross and Wadge, 1988; Wang and Howarth, 1989; Wang and Howarth, 1990). The approach in this study was to use the Hough transform in a special version that is more sensitive to grey level continuity in digital aerial photographs and satellite images. After the detection of strong linear features on the image, we examined the data available at different scales from a level of ground information compared to that obtained from aerial photographs and satellite data. The lineament data studied comes from a range of fracture patterns from a simple two-set pattern to a more complex one having three or more sets. The Hough Transform Algorithm The Hough transform is designed to detect collinear sets of edge pixels in an image by mapping these pixels into a parameter space (the Hough space) defined in such a way that collinear sets of pixels in the image give rise to peaks in the Hough space (Ballard and Brown, 1982). The edge pixels processed by the transform include the digital approximations of the gray level gradient magnitude m and direction 0. A straight line can be characterized by its slope a and its perpendicular distance p from the origin. For the line with slope a that passes through the point (x, y) , we have p = x sin a y cos a. Note that p is a signed quantity: e.g., if a is in the fourth quadrant and (x, y) is in the first quadrant, p is negative. Suppose an image contains a long straight edge e having slope a and distance p from the origin. Then an edge pixel having coordinates (x, y) and lying on e should have gradient direction 8 perpendicular to a, i.e., a = 0 2 r l 2 . Thus, in terms of 0, Equation 1 becomes p = x sin (0 + ~ / 2 ) y cos (0 + ~ / 2 ) = + (x cos 8 + y sin 8). If we compute p for each edge pixel using Equation 1, e should give rise to a cluster of ( p , 0) values. Note that one needs compute only one p for each 8, depending on whether the positive gradient direction is 0 + d 2 or 8 m/2. Unfortunately, in practice one does not obtain "tight" Photogrammetric Engineering & Remote Sensing, Vol. 62, No. 5, May 1996, pp. 525-531. 0099-lll2/96/6205-525$3.00/0 O 1996 American Society for Photogrammetry and Remote Sensing