Automatic keyword recognition using Hidden Markov models

An algorithm for automatic recognition of keywords embedded in a poorly printed document is presented. For each keyword, two statistical models, named Hidden Markov Models (HMMs), are created to represent the actual keyword and all the other extraneous words, respectively. Dynamic programming is then used to measure the Bayesian distortions of an unknown input word with respect to the two models and making a maximum likelihood decision. The HMM facilitate a nice 'elastic matching' property which makes the recognizer tolerant of highly deformed and noisy words. The system is shown to be robust, failing only when the levels of degradation are quite severe.