AN ADAPTIVE MULTI-DIMENSIONAL SPACE MODEL FOR VISUAL RECOGNITION USING VERBAL CONCEPTS

The definition and manipulation of complex visual scenes are a continuing problem in computer-aided design, computer-aided manufacture (CAD/CAM) and in computer generated imagery. Usually the human-to-computer interaction is done with a specialized set of commands, an arcane set of mnemonics or a specialized language. In all these cases extensive training is required for effective image generation; also novice users are often scared away from using computed imagery for aiding designs. This treatise describes a method of generating and handling of computer generated imagery; the method is embodied in a computer program which accepts a natural language character stream as its input and produces actions on the computer display screen. The images are handled as points in a multidimensional concept space, and natural language descriptions of three-dimensional solid objects are displayed as visual images by the computer's display subsystem. This process forms the basis of a linguistic model for graphic entities. An object or a grouping of objects is mapped as a point or a set of points in a multi-dimensional space and properties of the object are changes in the coordinates in that space. In this system, new concepts may be input in a conversational manner based upon combinations of previously defined concepts. These new concepts are then reevaluated and may be subsequently used to form higher level concepts at the operator's discretion. New OBJECT concepts exist as a hierarchical structure of points in a multi-dimensional feature space, based upon more primitive OBJECTS. MODIFIERS are used to change their appearance by the vector displacements of these points. The concept space model is a quantitative medium for the representation of visual and verbal concepts and is compatible with accepted qualitative linguistic theories.; A portion of the model is implemented as a natural language based computer graphics language. This program uses a 'Structured' parser which incorporates a flexible strategy to transform typed English commands into the appropriate multi-space conceptualization for direct execution by a computer graphics system. A sample dialogue with the program is also discussed.