Adaptive Correction of Errors from Recognized Chinese Ink Texts Based on Context

Chinese ink texts can not be converted into encoded texts until their writing characters are correctly recognized. There are many errors in recognized Chinese ink texts even importing language models because Chinese ink texts are free forms and mixed with other languages, as well as their Chinese characters have a large set and complex structures. Recognized writing characters may contain wrong language types, symbols, words, and word pairs. A direct selection and input approach based on context is proposed to adaptively correct theses errors. Each writing characterpsilas recognition candidates are fully visualized. Users can naturally and easily correct recognition errors with direct operations. Userspsila intensions are identified from userspsila gestures and objects invoked by them. Recognized Chinese ink texts can provide multi-levels of information after correction. We have conducted experiments using real-life Chinese ink texts and compared the proposed approach with others. Experimental results demonstrate that the proposed approach is effective and robust.