Texture segmentation as a first step towards archaeological object detection in high-resolution satellite images of the Silvretta Alps

Abstract

Since 2007, the Silvretta Archaeological Project in the high Alps on the Swiss-Austrian border has been investigating the prehistoric origins of alpine pasture economy. In an area of about 540 km2 more than 20 well-preserved archaeological sites associated with alpine pastoralism have been recorded, the earliest of them dating to the Iron Age (Reitmaier (ed.), 2012; Walser and Lambers, 2012). All of the ruined huts, cellars and livestock enclosures at these sites are visible on the surface and show a limited range of shapes and proportions. According to their function, all of them are located in open grassland. Based on this sample, we are currently developing methods to detect archaeological objects of the kind described above in high-resolution satellite images of our study area (Lambers and Zingman, in press). These methods are intended to assist archaeological survey in vast and/or difficult to access areas by screening large amounts of remotely sensed images in order to detect possible archaeological sites prior to fieldwork (Cowley, 2012). Our general approach aims at assessing the probability of the presence of objects of our interest based on geometric cues that can be automatically detected in the satellite and aerial images that we use. We here describe our general methodology and the first integral step constituting a new approach to texture segmentation.