Resumen
Automated detection of landscape patterns on Remote Sensing imagery has seen virtually little or no development in the archaeological domain, notwithstanding the fact that large portion of cultural landscapes worldwide are characterized by land engineering applications. The current extraordinary availability of remotely sensed images makes it now urgent to envision and develop automatic methods that can simplify their inspection and the extraction of relevant information from them, as the quantity of information is no longer manageable by traditional ?human? visual interpretation. This paper expands on the development of automatic methods for the detection of target landscape features?represented by field system patterns?in very high spatial resolution images, within the framework of an archaeological project focused on the landscape engineering embedded in Roman cadasters. The targets of interest consist of a variety of similarly oriented objects of diverse nature (such as roads, drainage channels, etc.) concurring to demark the current landscape organization, which reflects the one imposed by Romans over two millennia ago. The proposed workflow exploits the textural and shape properties of real-world elements forming the field patterns using multiscale analysis of dominant oriented response filters. Trials showed that this approach provides accurate localization of target linear objects and alignments signaled by a wide range of physical entities with very different characteristics.