High performance shape recognition using a novel multiple expert recogniser configuration

This paper presents a novel multiple expert configuration for efficient shape recognition using a priori knowledge about the target database. It is demonstrated how a generalised configuration comprising an integrated set of multiple independent classifiers can be implemented to enhance the overall recognition performance of the system. Since the super-class structure is known in advance, further refining of the classification process can take into account individual structures common to all the classes building up that super-class and the decision making process becomes easier. Moreover, as the number of target classes are reduced, the classification algorithms are able to separate the shapes by doing efficient clustering in the corresponding feature spaces. It is also demonstrated that such a configuration helps in optimising multiple isolated classifiers in the framework of the overall configuration for particular task domains such as automated inspection and security applications