Recognition Of Printed Text Based On Dhmm

A novel printed character recognition method based on Degraded Hidden Markov Model is proposed in this dissertation, which can be applied to engineering.For the construction of character is invariant relatively, the traditional structural recognizing method, which adopted usually, gets good recognition rate. In the aspect of feature vector picked up, the thinning algorithm was introduced usually to get the strokes in most cases in structural method, but the thinning can cause the mistakes in getting the strokes, then it cause the syntax deduce mistakes, so it can drop down the recognition rate, and the method which used to revise the thinning mistake is so complex. In the application of HMM on character recognition, the left right model structure was adopted not only in on-line but also off-line character recognition, but it can not fit the feature vector which proposed in this dissertation.By analyzing the structure of the character, a novel feature vector is designed in this dissertation, which called observe sequence vector. the vector fully represents a character, and can convert the 2D character information into 1D signals, the vector reserves the character's geometrical information. Based on the traditional HMM, we propose a new classifier—DHMM (Degraded Hidden Markov Model) to deal with 1D sequence. A different model of state's transition is adopted, and the model structure is ergodic.The result illustrates that the character type numbers have little effect on our method, and the DPIs which reach some resolutions have the same effect too. The experiment probe that the vector and the classifier is fit for printed character recognition well.