A bottom-up and top-down approach to using context in text recognition

摘要: Existing approaches to using contextual information in text recognition tend fall into two categories: dictionary look-up methods and Markov methods. use transition probabilities between letters represent a bottom-up approach context which is characterized by being very efficient but exhibiting mediocre errorcorrecting capability. Dictionary methods, on the other hand, constrain choice of letter sequences be legal words top-down impressive error-correcting capabilities at stiff price storage computation. In this paper, combined algorithm proposed. Exhaustive experimentation shows that achieves capability half cost.