High-level decisions in belief networks based 3-D object recognition

An important function in high-level computer vision is concerned with model representation and the selection of suitable sets of features for constructing each model. In this paper, we present a set of form invariant inferences following the basic concept of viewpoint invariance and a set of invariant analysis. It is a key component for representing the general forms of 3-D objects. We focus on using simple belief networks in representing the general forms and propose several geometric constraints to reduce the form combinations by chances. It is also greatly reducing the complexity of belief calculations. The invariant inferences and geometric constraints provide high-level decision rules for the belief networks based 3-D object recognition