Regarding computer vision as optimal decision making under uncertainty, a new optimization paradigm is introduced, namely, maximizing the product of the likelihood function and the posterior distribution on scene hypotheses given the results of feature extraction. Essentially this approach is a Bayesian formulation of hypothesis generation and verification. The approach is illustrated for model-based object recognition in range imagery, showing how segmentation results can optimally be incorporated into model matching. Several new match criteria for model based object recognition in range imagery are deduced from the theory.
展开▼