LARGE-SCALE SIMULATION STUDIES IN IMAGE PATTERN-RECOGNITION

Many obstacles to progress in image pattern recognition result from th e fact that per-class distributions are often too irregular to be well -approximated by simple analytical functions. simulation studies offer one way to circumvent these obstacles. We present three closely relat ed studies of machine-printed character recognition that rely on synth etic data generated pseudorandomly in accordance with an explicit stoc hastic model of document image degradations. The unusually large scale of experiments-involving several million samples-that this methodolog y makes possible has allowed us to compute sharp estimates of the intr insic difficulty (Bayes risk) of concrete image recognition problems, as well as the asymptotic accuracy and domain of competency of classif iers.