A two-level model-based object recognition technique

Object recognition is the problem of detecting the presence and determining the pose of a set of known objects in given images. For some applications, the known objects may be composed of identical components. The relations among these components can be exploited to improve recognition accuracy. This paper introduces a two-level model for object recognition which reduces redundant work due to objects with identical components by explicitly specifying the components and the relations among them. Using automatic analysis of music scores as example, an empirical study is presented, demonstrating the effectiveness and properties of the technique