FUZZY-SET-BASED HIERARCHICAL NETWORKS FOR INFORMATION FUSION IN COMPUTER VISION

A new methodology based on fuzzy-set-theoretic connectives to achieve information fusion in computer vision systems is introduced. The proposed scheme may be treated as a neural network in which fuzzy aggregation functions are used as activation functions. The scheme involves generating degrees of satisfaction (memberships) of various decision criteria and aggregating the memberships in a hierarchical network. The nature and the parameters of the aggregation connectives are learnt through neural network training procedures. We present techniques to determine the structure of such networks when this structure is only approximately known. These techniques also provide a mechanism for selecting powerful features and discarding irrelevant features via the detection of redundancies. Another attractive feature of the proposed approach is that the networks that result after training can be interpreted as a set of "rules" that may be used for decision making. Applications of this method in various computer vision tasks are presented.