A Study On Recognition For Rotated Isolated Online Handwritten Chinese Character

As the personal digital assistants with pen-based input is increasingly popular, handwriting input method as one of the important means for interaction of human and computer becomes more and more important. However, most Chinese character recognition systems currently in use impose a restriction on users with a writing box, and they require no distortion for the input character, especially no rotation, or poor recognition accuracy will be expected. In order to construct an unconstrained online handwritten Chinese character recognition system, it is necessary to develop the rotation-invariant online character recognition technology.In this paper, we work on research of recognizing technologies for rotated isolated online handwritten Chinese character. Through study on preprocessing method of input Chinese character, we propose a novel preprocessing step, called rotation-normalization, to achieve recognition for rotated isolated online handwritten Chinese character. Guided by the idea of rotation-normalization, we present and analyse two algorithms to implement rotation- normalization as follow:1. Tilt-?Normalization MethodTilt-?Normalization Method includes three steps, skew angle rectification according to the starting point(SAR-SP) , angle readjustment using direction axes(AR-DA) and Training classifier using artificially rotated samples(TC-ARS), this method implements rotation-normalization based on the position of the starting point and the information of writing direction of the input character.2. Suspension MethodBased on the physical phenomenon of gravity suspension, Suspension Method takes the input character as object that has mass, and then suspends it to implement rotation-normalization.The experiments show that our proposed tilt-normalization method and suspension method are simple but greatly improve the recognition accuracy for rotated isolated online handwritten Chinese character, and they don't require extra memory space, therefore, they are practically useful on handheld equipments.The proposed novel preprocessing step rotation-normalization is generally applicable, and it provides basic thought and research direction for the further study of recognition for rotated isolated online handwritten Chinese character.