A Uniied Network-based Approach for Online Recognition of Multi-lingual Cursive Handwritings

Although several studies have focused on recognition of individual language, no attempt has been seriously made for online recognition of handwritten script in multiple languages. In this paper, a network-based approach is proposed for recognizing sequences of words in multiple languages. Viewing handwritten script as an alternating sequence of words and interword ligatures, a hierarchical hidden Markov model(HMM) is constructed by interconnecting HMMs of ligature and word of multiple languages. In turn, those HMMs are constructed from HMM of lower level components. Given such a construction, recognition corresponds to nding optimal path in the network which can be searched eeciently by Viterbi algorithm. Although combining component languages, recognition accuracy of each language drops negligibly little, and recognition result of intermixed usage is acceptable.