Evaluation of combination methods

This paper presents a new approach to evaluate the performances of combination methods, which takes into account both the recognition rates of the experts combined and the correlation among them. At the purpose, a suitable estimator of correlation is defined. Two combination methods have been considered: majority vote and Dempster Shafer. A statistical test, based on the analysis of variance, has also been used to infer some interesting considerations on the behaviour of combination methods. The paper shows how the proposed approach allows the selection of the best combination method for each set of experts.

[1]  Ching Y. Suen,et al.  Complementary algorithms for the recognition of totally unconstrained handwritten numerals , 1990, [1990] Proceedings. 10th International Conference on Pattern Recognition.

[2]  Ching Y. Suen,et al.  Computer recognition of unconstrained handwritten numerals , 1992, Proc. IEEE.

[3]  Kyusik Chung,et al.  A systematic approach to classifier selection on combining multiple classifiers for handwritten digit recognition , 1997, Proceedings of the Fourth International Conference on Document Analysis and Recognition.

[4]  Ching Y. Suen,et al.  Historical review of OCR research and development , 1992, Proc. IEEE.

[5]  Adam Krzyżak,et al.  Methods of combining multiple classifiers and their applications to handwriting recognition , 1992, IEEE Trans. Syst. Man Cybern..

[6]  Hiroyuki Shimizu,et al.  Complementary classifier design using difference principal components , 1997, Proceedings of the Fourth International Conference on Document Analysis and Recognition.