Image segmentation for the automatic recognition of historic masonry walls

This study approaches the potential of image recognition based on artificial intelligence to automate the construction analysis of stone walls, as support to the typological study of historic masonry. Santo Stefano di Camastra, Sicily, was used as case study. A set of 60 photos, each representing one of the visible walls in the historic centre, were assessed in six aspects of image quality and three aspects of wall complexity. Each photo was segmented through the open access tool Meta AI and a score was assigned to the quality of segmentation. The association was analyzed between the quality of segmentation and the variables on image quality and wall complexity, considered individually or in linear combinations. Somers’D index was used. The study suggests that image quality is partially associated with the result of automatic segmentation through the presence of other building components and external elements in the picture, and the focal distance of photography. Association was not observed between segmentation and construction features of masonry, while the relationship is strong with features describing the appearance of wall surface.