Description and recognition of animal silhouette image using ellipsoid‐expansion
Junichi Hara, Hirokazu Kato, Seiji Inokuchi · Systems and Computers in Japan · 1992
Abstract Although objects in most computer vision have been assumed to be “solid bodies,” this assumption is not always true in practice; for example, a horse can be an object. It has been known that humans recognize such an object by dividing it into several parts, each of which has a recognizable feature. This process is important also for computer processing. This paper proposes pattern recognition of a silhouette of an animal using the relative position of each part of its body. The process is to segment each region of the silhouette and to describe the characteristic of each part and the relationship between a part and its adjacent part. A new method, the “ellipsoid‐expansion method,” based on an ellipsoid fitting, is proposed to segment a silhouette. This method uses an ellipse which is fitted to a region by expanding the ellipse and by producing a reaction when the expanded ellipse exceeds the contour of the region. This method is not influenced by details of the shape of the object and withstands shifts, change of scaling, and rotation of the shape. Recognition of each part of the silhouette of an animal is tested by identifying the description of each part of a basic model of the animal and the description of its silhouette. Examples of identification of animals are carried out using some simple cases.