Automating detection of landscape patterning: a similarity-based approach

Landscape patterns, which are the product of composite historical landscape engineering, such as land divisions and field systems, are particularly suited to be automatically identified on remote sensing imagery using methods for computer-aided identification. A variety of approaches in Pattern recognition, such as feature extraction methods, provide the opportunity to implement new routines able to reveal possible anthropogenic landscape components based on recognition of recurring patterns and regularities in aerial, satellite and Lidar datasets. Shortcomings and deficiencies of former methods based on simple automated pattern matching are thus circumvented, enabling operators to identify landscape patterns designed by a range of environmental or manmade elements. This paper will expand on the implementation of this approach over the landscape surrounding the Roman city of Aquileia (Italy), shaped by an extensive centuriation, the Roman system of land subdivision into large regular plots allotted to colonists, starting from the end of the 2nd century BC. Under the umbrella of the VEiL Project, the method is utilised to automate procedures of similarly-oriented linear feature detection in the remote sensing imagery and to supplant object detection procedures based on individual examination and interpretation.