On creating averaging templates

Dynamic Programming is normally used in creating averaging templates. However, it takes both time and memory. In this paper, we present linear averaging and dynamic local averaging as two alternative approaches. A recognition experiment was also conducted for performance comparisons. It was found that both proposed techniques not only achieve a slightly higher accuracy than dynamic averaging, but also save time and memory. The effect of length normalization on dynamic averaging and dynamic local averaging was also studied and was found not to be a significant factor.