A system for recognizing Vietnamese document images based on HMM and linguistics

The authors present a system for recognizing Vietnamese document images and propose a method to increase the accuracy for this system. Based on features of the Vietnamese language, we can minimize the number of characters and integrate spell-checking in the recognition process. We also explain how to combine HMMs and our method in the recognition systems. Finally, based on statistical models for word frequency, a dictionary of Vietnamese word frequency was built to predict the next words to be recognized and to aid in post processing. The performance of the proposed approach was evaluated on Vietnamese literature from 1990 to 1997 with a total of 3469518 words (about 16866511 characters). Experimental results show that our method was effective.