A method to detect distinctive features from boundary data
Kentarou Yamagishi · Systems and Computers in Japan · 1991
Abstract Recently, attempts have been made regarding automatic learning of models of objects with two‐dimensional shapes from their multiple number of images and recognizing objects using pregiven models by partial features of shapes. Both are based on a principle of extracting the local features of a figure and using them as distinctive features for both learning and recognition. This paper proposes a method of feature extraction from a boundary pattern intended to be applied for such purpose. A hinge articulation is considered as a basic component of two‐dimensionally shaped objects, and for each one of the hinge representative shapes—a circle, linked bars and crossed bars—a simple but robust detection method is presented. Since extraction of line segments is necessary as a preprocess for detecting linked and crossed bars, a line‐segment detection method applying Hough transformation is developed for boundary patterns.