A Neural Network-Hidden Markov model hybrid for laser etched characters recognition

A recognition method of laser etched characters based on Hidden Markov models and Neural Network is applied, which the structural properties and the relation of metal label characters are analyzed in detail. The endings, three crossing points, four crossing points of the characters are extracted, and it is improved on the method of extraction of three crossing points. A neural network is used to estimate probabilities for the characters depended on the structural properties, then deriving the best word choice from a sequence of state transition. It is shown in test that the proposed method can be used to recognize the etched characters on metal label.