Interpretation of Ambiguous Zone in Handwritten Chinese Character Images Using Bayesian Network

Interpretation of ambiguous zone is an essential step to recovering dynamic information from handwritten images, which can be seen as to deduce the original motion intention of the writer at the intersection areas. This study presents a novel method to interpret ambiguous zones by constructing a Bayesian belief network. In the initial phase, a graph is built to model the character and several sample points are extracted from each sub-stroke. In the interpreting phase, each pair of sub-strokes is characterized in terms of the comparison of orientation, width, and curvature. Finally, a Bayesian belief network is established to determine the continuous pairs. A series of experiments are conducted on test samples collected from a standard handwritten Chinese text database, and the results show that the proposed method can interpret ambiguous zones effectively.