Fuzzy control structures in multiple parameter systems: an application in handwritten address interpretation systems

Abstract

Most practical software systems are a conglomerate of many modules, where each module has its own parameters that control the accuracy of the module. While each individual module can be optimized by tuning the relevant parameters, it is a non-trivial task to optimize the entire system. When the number of modules and the parameters are few, manual choosing of all parameters is possible by a "trial and error" mechanism. However, when the modules are many, other methods have to be adopted. We use fuzzy set technology for the purpose in the particular application of a Handwritten Address Interpretation System. We use a fuzzy measure based on the topology of the "blind" set of handwritten addresses and a predetermined neighborhood to construct the fuzzy membership function. This allows a nonlinear partitioning of the results to maximize the correct rate. On a test set of 10000 images, the fuzzy methodology accomplishes 5% higher accuracy.