Word recognition from sparse graphs of letter candidates using wildcards and multiple experts

Variability in handwriting styles suggests that many letter recognition eng ines cannot correctly identify some hand-written letters of poor quality at reasonable computational cost. Methods that are capable of searching the r esulting sparse graph of letter candidates are therefore required. The meth od presented here employs 'wildcards' to represent missing letter candidate s. Multiple experts are used to represent different aspects of handwriting. Each expert evaluates closeness of match and indicates its confidence. Exp lanation experts determine the degree to which the word alternative under c onsideration explains extraneous letter candidates. Schemata for normalisat ion and combination of scores are investigated and their performance compar ed. Hill-climbing yields near-optimal combination weights that outperform c omparable methods on identical dynamic handwriting data.