Bayesian networks and utility theory for the management of uncertainty and control of algorithms in vision systems

An Image Understanding (IU) system should be able to identify objects in 2D images and to build 3D relationships between objects in the scene and the viewer. The system presented here has a control structure for general purpose image understanding that addresses both the high level of uncertainty in local hypotheses and the computational complexity of image interpretation. The control of vision algorithms is performed by an independent subsystem that uses a set of Bayesian networks and utility theory to compute the expected value of information provided by alternative operators and selects the ones with the highest utility value. Each operator has a cost, which is related to the algorithm complexity associated with the operator. The cost of each operator is considered during the operator\u27s selection process. This control structure was implemented and tested on several aerial image datasets. The results show that the knowledge base used by the system can be acquired using standard learning techniques and that the value-driven approach to the selection of vision algorithms leads to performance gains. Moreover, the modular system architecture simplifies the addition of both control knowledge and new vision algorithms