Abstract
Image recognition technology using computers has been active recently. In image recognition, it is necessary to automatically extract elements such as straight lines, circles, numerical letters, and symbols from a picture. There are many cases in which the extraction of such elements is not easy because some elements overlap. This paper describes a method for extracting individual overlapping elements from a binary image. This method can extract an element which has a constant shape with arbitrary size or rotation angle, or with both an arbitrary size and an arbitrary angle. The method uses a single key point and several feature points on the model element to be extracted. The extraction is carried out by checking if the pixels corresponding to the feature points are present, with each pixel regarded as a key point of the model. This paper describes the algorithm, a method of extracting some important straight lines and circles in the picture using the algorithm, and experimental applications of the method to mechanical drawings. The results of the experiments demonstrate the successful extraction of straight lines, circles, arrows, and alphanumerics, and this confirms the usefulness of the proposed method.