Multiple classifier fusion using the fuzzy integral.

Date of Award


Publication Type

Master Thesis

Degree Name



Electrical and Computer Engineering

First Advisor

Ahmadi, M.


Engineering, Electronics and Electrical.



Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.


Fusion of multiple classifier decisions is a powerful method for increasing classification rates in difficult pattern recognition problems. Researchers have found that in many applications it is better to fuse multiple relatively simple classifiers than to build a single sophisticated classifier to achieve better recognition rates. Ideally, the combination function should take advantage of the strengths of individual classifiers and of all possible subsets of classifiers, avoid their weaknesses, and use all the dynamically available knowledge about the inputs, the outputs, the classes, and the classifiers. Automatic reading of handwritten numerals is a difficult problem because of the great variations involved in the shape of the characters. In this thesis an evidence fusion technique, based on the notion of fuzzy integral is utilized to combine the results of different classifiers and realize a robust algorithm for high accuracy handwritten numeral recognition. Both source relevance as well as source evidence are utilized to achieve significant enhancements. The most important advantage of this system is that not only is the evidence combined but that the relative importance of the different sources is also considered. Various conventional and fuzzy integral based fusion methods are explained in detail and experimental results obtained are compared. A method is introduced to improve the fuzzy densities of the classifiers which would improve the fusion results. In this method we use the correction factors obtained from the performance matrices to alter the initial fuzzy densities. Experiments on handwritten numeral recognition are described and compared. These experiments show that very low error rates can be achieved by fusing several low performance classifiers.Dept. of Electrical and Computer Engineering. Paper copy at Leddy Library: Theses & Major Papers - Basement, West Bldg. / Call Number: Thesis1999 .B45. Source: Masters Abstracts International, Volume: 39-02, page: 0558. Adviser: M. Ahmadi. Thesis (M.A.Sc.)--University of Windsor (Canada), 1999.