Abstract :
In this paper, we argue that in order to be able to build meaningful image filters, we have to have a world model, a description of the world that we are dealing with. This world model includes a model of the scene, a model of the eye (camera), and a model of the illumination. Secondly, using this description, we have to recognize objects in the scene which match this description before we can remove them, that is, filter them out. Thirdly, we show that the strategy of sequencing the procedures (deciding which object in the scene is recognized when) is guided by the world model and the visibility of objects on the scene. Since the filtering process involves recognition of the visual concepts of objects, we call it conceptual filtering. Similarly, since the guided strategy of the scene analysis, depends on partial recognition of the scene, we call it conceptual focusing. The preceding points are demonstrated by a concrete example of computer recognition of bridges, rivers, lakes, and islands from ERTS satellite pictures.
