Abstract
We investigate the combination of Type-III classifiers using the Dempster-Shafer Theory of Evidence. Various methods of building BPA?s for each classifier using both "global" and "local" classifier information are explored. We propose modifications to two established BPA-computation methods to make them better suited for combining Type-III classifiers. We also show the effectiveness of using compound hypotheses when a classifier cannot confidently choose between the top two returned classes. Experimental tests demonstrate the superiority of some of the approaches proposed here on the numeral recognition problem when combining three different character recognizers with Type-III classification engines.