Retrieval of cursive chinese handwritten annotations based on radical model

Abstract

Cursive Chinese characters are analyzed using a semantic matching process whereby radicals within the character are first extracted and used to reduce the search space of the full lexicon to only those characters containing the matching radical. In performing the radical extraction, the input character is normalized and segmented into strokes that are in turn organized based on stroke up/down information and local maxima and minima information. Obscure breakpoints and connecting strokes are removed in the process. Dynamic program matching is then performed on a stroke basis in which stroke substitution costs are assessed on a point-by-point basis through a variety of techniques, including tangent vector analysis, center relationship assessment and starting point/ending point assessment. Dynamic programming costs are normalized based on the length of the reference radical and location dissimilarities are removed.