A Graph-based Object Structural Representing Method

Most objects can be divided into several components which we call "entities". That means an object is usually a structural combination of entities which are arranged in some patterns associated with this kind of object. To get the representation of object structure is a challenge in the field of cognition. This paper presents a methodology for object structural representation which can be used in object semantic modeling, object recognition, image understanding and so on. This methodology is based on a graph approach which is able to record the local and global characters of object, measure the similarity between two objects and extract the common structures of several objects. The local features of object are based on shape, location and size. And the global features correspond to the topological relationships between the object's entities. Based on this representation modal, the synthetic object classification task is undertaken by using graphs comparing and nearest neighbor algorithm. Experiments are done on some 2-D CAD objects which are assembled with geometric entities. And the results show the rationality and effectiveness of this methodology on object classification.