Introduction

In the previous chapter, we have seen that a $k$-means like algorithm can help us to sample a sequence of active appearance models into a sequence of prototypes. In this chapter, our aim is to be able to learn these sequences in order to predict prototypes in an unseen sequence. The learnt model should also be able to generate a synthetic sequence of prototypes. In this ideas in mind, the variable length Markov model (VLMM) has been investigated.