Augmenting reality in the tasks of classifying objects in aerospace images

Abstract

Abstract The problem of increasing the reliability of object recognition in aerospace images consists in the insufficient quality of image segmentation based on pixel colour characteristics and boundary shape descriptors. The paper proposes an approach based on description logic with an extension on graphs. The essence of the approach is the searching for relevant cases in the ontological base by calculating the degree of similarity and identifying differences of attributed graphs of cases. With the help of retrieved cases, the object structure and the boundary shape are refined. After that, objects can be augmented with subject domain information stored in the ontological base. Representation of the structure of cases in the form of attributed relational graphs allows avoiding the rapid growth of the case base, since the similarity of connected subgraphs is generally sufficient. The experiments are conducted on urban area aerospace images.