A queueing problem in pattern recognition

Patterns arrive at random instants to a sequential recognizer and possibly form a queue. Each pattern is recognized through one or more consecutive stages following a definite discipline; the times spent by a pattern in the various stages are assumed to be independent random variables. The resulting model of the recognizer is a Markovian queueing system with a service discipline which aims to minimize both the queue-size and the error probability. The traffic intensity of the patterns and the probabilities of entry in the various stages are the parameters which determine the equilibrium probabilities of the system. From these the error probability and other characteristics of interest follow. Afterwards the hypothesis of the independence of the service times is removed other special cases are also considered.