An Algorithm of General Closeness Degree Based on Evidence Theory and Its Applications in Fuzzy Pattern Recognition

Abstract

With the different features and applications for each of the conventional closeness degree algorithms, the mistake will be brought in pattern recognition if it was not selected correctly. Based on the combination rules in D-S evidence theory, a new algorithm was proposed by balance weighted between two conventional closeness degrees. The calculated data corresponding each conventional closeness degree was weighted as independent evidence; the weights were adjusted adaptively in terms of the distribution of pattern eigenvalue. The new algorithm combines the membership degree with closeness degree and can calculate the pattern eigenvalues represented by general real number, sections and fuzzy sets on the real number domain, which integrates the direct method and indirect method in pattern recognition. In every item of evidence there are weighted factors and the weights can be adjusted adaptively, so the recognition exactness rate can be improved. The validity of this algorithm was confirmed by two examples.