A Study Of The Hidden Markov Models Of Pattern Recognition

Pattern recognition is ability which is to input original date and in terms of the category to take corresponding actions, and is a kind of complex processing mechanism. Generally speaking, pattern recognition mainly includes three subjects: isolation, parameterized and the model used for training and recognition.The Hidden Markov Model is one of the important models in the pattern recognition. They are the double stochastic process and which consist of a hidden Markov process and a displayed random function set, also, they can provide important model reference to process pattern recognition problems. They were brought forward by Baum and his colleagues in the late sixties and early seventies. Then, they have been widely used in speech recognition, statistical language learning, sequence symbol recognition, optical recognition and image processing etc. There are three problems needed to be solved by using hidden Markov models, which are valuating, decoding and recognizing. The answers of these three problems consist of the theory of hidden Markov models.These questions are discussed in the article:1. In pattern recognition, there are three basic operations, they are pretreatment feature extraction, classification. But with the time of processing, the moment events that occurred in the front will suffer a moment of the events of direct effect. This paper introduces first-order conditions corresponding actions, apply the models to second actions and illuminate the principle and achieve of second-order Hidden Markov Model.2. The new algorithm combines the advantage of entropy model, which can integrate and process rules and knowledge efficiently to deal with the mass information process. This algorithm uses the sum of all features with weights to adjust the transition on parameters in hidden Markov model for information extraction.3. According to use large study records as training sets, the new algorithm combines the entropy model and the improvement of assumptions of the state transitions and the output observations in hidden Markov models to establish hidden Markov model. Experimental results show that the new algorithm improves the performance in data processing, calculation and accuracy. 4. This paper introduces the probability of data vectors at any given moment not o- nly relies on the present moment sates of the system, but also relies on the structure of the Mixed Hidden Markov Models in the premise of data vector at the moment before the system. It discusses the forward-backward algorithm of the new model, and deduces and proves the revolution formula of parameterαk in this model.