An approach to the recognition of contours and line-shaped objects

Abstract Some methods implemented for printed circuit (PC) board inspection have been developed and applied to line analysis. The idea of calculation of the number of objects within a circular operator field, developed by Danielsson and Kruse (Computer Graphics and Image Processing 4, 1979) for distance-checking algorithms, has been elaborated for line recognition. An edge-preserving technique with a rotating neighborhood has been implemented for texture recognition. The local properties of the operator field with the rotating neighborhood have led to formulating and proving a basic condition on representing images. This condition relates conformity of recognition results with group theory. The results of line inspection by means of a circular operator following the line are represented as graphs. The nodes of these graphs are labeled with names of features of the image (e.g., of the line followed) and the arcs of these graphs are labelled with the names of relations on these features. As soon as the names of these features and relations have been expressed, usually through words and phrases, a description of the image can be made with the aid of expressions of natural language, and then converted to graph form.