Abstract
A method for segmentation of quantized patterns from a standpoint of connectivity is proposed. A pattern consisting of black points connected to each other is separated as a unit. These points are labeled during a scan of a picture. By an algorithm shown in this paper the same label is given to the points that join together. After one scan of the picture, a pattern is separated from the others by making use of the labels given to the points. The use of this method makes it also possible to count patterns. The number of labels used indicates how many patterns are contained in the picture. An improved rather noise-free method is also propoosed, in which a pattern is regarded as compsed of segments instead of points. The method enables us to pick up information while scanning for the segmentaion, which is necessary to extract everal kinds of features from each pattern with ease. These are global features, such as a circumscribed rectangle, center of gravity and axes of symmetry, and partial ones, such as corner points, loops and end points. The information is available for discriminating patterns or for adapting the segmentation algorithm to objects.
