Research On Spatial Relation Description For Two Dimensional Images

Space plays a fundamental role in human cognition. In everyday situations, it is often viewed as a construct induced by spatial relationships, rather than as a container that exists independently of the objects located in it. Spatial relationships, therefore, have been thoroughly investigated in many disciplines, including artificial intelligence, cognitive science, psychology, linguistics, geography, graphics and so on. In image understanding and related fields, understanding the spatial organization of regions in images and building the linguistic descriptions of spatial relation automatically are two important tasks, and these two aspects constitute the two levels of the study of spatial relation description. For the subject of spatial relation description, this paper thoroughly studies the fundamental theory of spatial directional relation, the formal model of spatial directional relation and linguistic descriptions of spatial relation. The main achievements are concluded as follows:(1) The perfect fundamental theories of spatial directional relation are proposed. Firstly, several clear definitions, including spatial directional relation, types of spatial object, reference frame of spatial directional relation, observation angle and basic direction, are proposed, and they lay the foundation for spatial direction relation research. Secondly, properties and characteristics of spatial directional relation are analyzed, and they provide the basic principles for the formal model of spatial directional relation. Thirdly, two classification methods of spatial directional relation are proposed, and they provide related basis for the formal model of basic directional relation and special directional relation and the linguistic description method based on detailed spatial relation.(2) The influence factors of spatial directional relation are proposed, and existing formal models of spatial directional relation are analyzed. Firstly, through the establishment of the graphic library of spatial directional relation, the factors that affect spatial directional relation are proposed and analyzed, and they provide theoretical basis for the formal model of spatial directional relation. Secondly, the characteristics of existing formal models of spatial directional relation are analyzed based on the fundamental theories and the influence factors of spatial directional relation, the results of analysis show that histogram model has fully taken into account orientation angle, shape, and relative distance of spatial objects to impact on representation of spatial directional relation. In this paper histogram model is defined basic model of spatial directional relation for its higher accuracy and applicability.(3) The perfect model system of spatial directional relation is proposed. Firstly, the visual area histogram model in observer reference framework is proposed, and it uses visual area between objects as the object of study based on visibility theory and effectively solved the problems of description in observer reference framework. Secondly, a new quadtree histogram is proposed, and it uses quadtree arithmetic to divide spatial objects into a number of simple and regular sub-objects based on two factors affecting the integration of graphics. Quadtree histogram is calculated based on the sub-objects, and it can ensure high accuracy and low computation. Thirdly, a new fuzzy judgment approach of the basic directional relation is proposed, and it using the availability degree and the certainty degree of directional relation to resolve the problems of existing judgment approaches of the basic directional relation. Finally, a new fuzzy judgment approach of the special directional relation is proposed, and it can resolve the problems of existing judgment approaches of the special directional relation. The experimental results show that the visual area histogram model and the quadtree histogram model are in better harmony with human spatial perception.(4) The linguistic descriptions approaches of spatial relation are proposed. Firstly, the common syntax patterns and fine-grained word-formation patterns are found by semi-automatic pattern extraction method. Secondly, several directional terms, distance terms and degree adverbs are automatically generated based on histogram models and fuzzy rules. Thirdly, the linguistic descriptions of spatial relation are automatically generated based on syntax patterns and fine-grained word-formation patterns in inner reference frame and outer reference frame. Finally, a linguistic descriptions approach of detailed spatial relation is proposed based on detailed spatial relation in different reference frames, and it can resolve the problems that the descriptions are not detailed enough in some topological conditions. The experimental results show that the linguistic descriptions approaches in this paper can accurately generate linguistic descriptions of spatial relation which are basically accordant with actual communication habits of human.