Multi-classifiers neural network fusion versus Dempster-Shafer´s orthogonal rule

This paper describes a system managing data fusion in the pattern recognition (PR) field. The problem is seen from the multi decisional point of view. Several modules´ classification specialized on specific features sub-spaces allowing the cooperation of different classification techniques. The use of neural networks for heterogeneous, incomplete and noisy data fusion permits the specification of the fusion module for a given application. Experiments are compared with fusion performed by the Dempster-Shafer´s orthogonal rule, proving the performances of such a system