Hierarchical Evidential Reasoning Networks for Object Recognition in an Outdoor Scene

A new methodology for classification of objects and categorisation of concepts is proposed. Fuzzy concept or category can be defined by a set of examples. Evidence is collected from all the features in each example in the example set. If a feature of an example appears to exist in an unknown object, it assumes that there is some support for the unknown object to be classified as the concept of that example. This method can be viewed as a hierarchical network. Unlike an artificial neural network, this method does not require the use of a large training set. The advantage for adapting this method to a hierarchical network architecture is to utilise its parallel architecture for hardware implementation. This methodology can be used in a vision system to provide a single coherent approach to scene analysis. Higher level fuzzy concepts can be defined consistently from pixel level to predicate level. A hierarchical Support Logic network was defined . It was applied to a set of outdoor scenes in order to demonstrate some of the principles behind the theory. Key word: Fuzzy Sets, Support Logic, Evidential Reasoning, Hierarchical Network, Scene Analysis, Classification and Categorisation