Methods for fusing multiple handwritten word classifiers are compared
on standard data. A novel method based on data-dependent densities in
a Choquet fuzzy integral is shown to outperform neural networks, Borda
and weighted Borda counts, and Sugeno fuzzy integral.