マルコフチェーンによる言語種類変更を用いたオンライン多言語文字認識

概要

論文の詳細を見る
Multi-lingual handwriting means the scripts written with more than one language. For on-line recognition of multi-lingual cursive handwriting, unified HMM (Hidden Markov Model) network-based approach was proposed. Since language mode changes even within a word, each language part in the recognition network were connected. Therefore network search space became huge and complex, which in turn, degraded the system performance. In this paper, Markov chain for language mode transition is proposed. Trends of language mode transitions are modeled by means of Markov chain and represented by transition probabilities. Therefore frequent language mode transitions can get merits by higher probability, as a result more accurate results are provided.
社団法人電子情報通信学会の論文
1997-03-06