Description logic on image graphs

In this paper, it's proposed to use the extension $\mathcal{ALC}(GI)$ of description logic $\mathcal{ALC}$ for the formal description and structural analysis of images. Concepts and the roles of $\mathcal{ALC}(GI)$ are interpreted on an image graph and its subgraphs. The model of image in the form of multi-layered attribute graph is presented. It contains a layer of color segments, a layer of boundaries and a layer of skeletons. Each layer is a planar graph, layers are linked by means of “ancestor-descendant” relations. The transition from the pixel representation of an image to the graph one allows us to increase the effectiveness of image analysis. Examples of terminological axioms that define structural elements of an image and letters composed of them, as well as the results of an experiment of recognizing letters in a cursive handwritten text are presented.