Dynamic fuzzy rule weight optimization for a Fuzzy Based Single-Stroke Character Recognizer

In this paper a dynamic fuzzy rule weighting method (DFW) combined with evolutionary optimization are presented for the formerly published Fuzzy Based Single-Stroke Character Recognizer (FUBAR) method. With the introduced rule weighting technique the consequent parts of the if...then... rules are calculated similarly to the original FUBAR method, but a dynamic fuzzy rule weight W_n([0,1]) described as a fuzzy set is applied to it in O_n·1/W_n(O_n) form, where O_n is the output of the rule. The membership functions of DFW-s are determined by bacterial evolutionary algorithm. The paper compares the results of the proposed new algorithm with other (formerly published) FUBAR algorithms and also with other commercial and academic single-stroke recognizers in terms of recognition accuracy and computational resources needed.