A Framework for Classifier Fusion: Is It Still Needed?

摘要

We consider the problem and issues of classifier fusion and discuss how they should be reflected in the fusion system architecture. We adopt the Bayesian viewpoint and show how this leads to classifier output moderation to compensate for sampling problems. We then discuss how the moderated outputs should be combined to reflect the prior distribution of models underlying the classifier designs. We then elaborate how the final stage of fusion should combine the complementary measurement information that might be available to different experts. This process is embodied in an overall architecture which shows why the fusion of raw expert outputs is a nonlinear function of the expert outputs and how this function can be realised as a sequence of relatively simple processes.