Pattern Recognition, 4th Edition

Recognition

Any path method

To start with, we will treat each of the M reference HMM models as a distinct class. The sequence of observations X is the result of emissions, due to transitions among the different states of the respective model. The problem then becomes a typical classification task. Given the sequence of N observations X : x¹, x², …, x^N, resulting from the unknown pattern, decide to which class it belongs. The Bayesian classifier decides in favor of pattern S* for which(9.21)and for equiprobable reference models (classes) this is equivalent to(9.22)where for convenience we have used S to denote the set of parameters describing ...