COMPARISON OF EDGE DETECTION AND HOUGH TRANSFORM TECHNIQUES FOR THE EXTRACTION OF GEOLOGIC FEATURES D. P. Argialas and O. D. Mavrantza Laboratory of Remote Sensing, School of Rural and Surveying Engineering, National Technical University of Athens, Greece, Heroon Polytechneiou 9, Zografos Campus, 157 80, Athens – rannia@survey.ntua.gr, argialas@central.ntua.gr Commission III, WG4 KEY WORDS: Remote Sensing, Geomorphology, Algorithms, Feature, Edge, DEM, Detection, Extraction, Automation ABSTRACT: Photointerpretation of geologic lineaments is a subjective process. Therefore there is a need for automation of lineament mapping using optimal edge detection techniques. Efforts made in this direction include the application of Sobel and Prewitt operators or directional detectors followed by edge linking techniques (e.g. the HOUGH Transform). It is difficult to choose optimal detectors, however, since the complex scenes portrayed on satellite images are strongly dependent on the radiometric and physical properties of the sensors and on the illumination properties and topographic relief of each scene. Therefore, the geographic region determines the “suitability” of an edge detector in geologic feature extraction. In this context, the objective of this work was the implementation, evaluation and comparison of selected optimal edge detectors and the HOUGH transform algorithm towards automated geologic feature mapping in a volcanic geotectonic environment. The test area was the Nisyros Island (Greece). A LANDSAT 5 - TM image and the DEM of the study area were geometrically corregistered with the scanned topographic map of the same area. The following edge detectors were applied and assessed on band 5 of the LANDSAT-TM image and the DEM, namely, (a) Canny, (b) Rothwell, (c) Black, (d) Bezdek, (e) Iverson-Zucker, (f) EDISON and (g) SUSAN. Modified versions of the HOUGH transform were additionally applied to these data. The resulted edge maps were quantitatively assessed with the use of evaluation metrics. Finally, the performance and behaviour of each algorithm for geologic feature extraction on the specific geotectonic terrain was investigated. 1. INTRODUCTION Geologic lineament mapping is considered as a very important issue in problem solving in Engineering, especially, in site selection for construction (dams, bridges, roads, etc), seismic and landslide risk assessment (Stefouli et.al., 1996), mineral exploration (Rowan and Lathram, 1980), hot spring detection, hydrogeological research, etc. (Sabins, 1997). Lineament photointerpretation is a quite subjective process, requires expertise, training, scientific skills and is time consuming and expensive. Therefore, the need arises for automation of photointerpretation in order to reduce subjectivity and to help the analysts. This can be achieved using computer-assisted techniques, e.g. image processing and analysis techniques, pattern recognition and expert systems. 1.1 Edge Detection Operators: Overview In image processing and computer vision, edge detection treats the localization of significant variations of a gray level image and the identification of the physical and geometrical properties of objects of the scene. The variations in the gray level image, commonly include discontinuities (step edges), local extrema (line edges) and junctions. Most recent edge detectors are autonomous and multi-scale and include three main processing steps: smoothing, differentiation and labeling. The edge detectors vary according to these processing steps, to their goals, and to their mathematical and computational complexity (Ziou and Tabbone, 1997). In the present work, only step edge detectors were examined, which can generally be grouped into three major categories: 1. Early vision edge detectors (Gradient operators, e.g. the detectors of Sobel and Kirsch). 2. Optimal detectors (e.g. the Canny algorithm, etc.). 3. Operators using parametric fitting models (e.g. the detectors of Haralick, Nalwa-Binford, Nayar, Meer and Georgescu, etc) (Ziou and Tabbone, 1997). 1.2 Edge Linking Techniques: Overview The HOUGH Transform is considered as a very powerful tool in edge linking for line extraction. Its main advantages are its insensitivity to noise and its capability to extract lines even in areas with pixel absence (pixel gaps). The Standard HOUGH Transform (SHT) proposed by Duda and Hart (Duda and Hart, 1972) is widely applied for line extraction in natural scenes, while some of its modifications have been adjusted for geologic lineament extraction purposes (Karnieli, et.al., 1996; Fitton and Cox, 1998). In the present work, a modified Hough Transform was applied in the satellite image and the DEM, namely the Fitton-Cox algorithm. This algorithm has successfully been applied to a sedimentary terrain covered with prominent joints. Its main advantage was the extraction of small line segments, which was controlled by the input parameters (Fitton and Cox, 1998). 1.3 Lineament Extraction and Mapping: Overview Concerning the semi-automatic and automatic lineament extraction, there are three main categories of processes: 1. The enhancement of geological line segments with the use of linear and non-linear spatial filters, such as directional gradients, Laplacian filters, and the Sobel and Prewitt operators (Morris, 1991; Mah et.al., 1995; Philip, 1996; Süzen, and Toprak, 1998), as well as morphological filters (Tripathi et.al., 2000).