| With the development of Mobile Internet and the increasing popularity of mobile terminals, character recognition technology has become matured. As an efficient input method on the mobile platform, handwriting has been accepted by more and more users. During the recent 30 years, although researchers make great efforts on enhancing the recognition rate, the errors are still not eliminated. In other aspects, constrained by the size limitation of the input area on the mobile device, the traditional input methods mismanagement the relationship between the input efficiency and the user experience.This thesis proposed a multi-modal error correction method for handwriting, and an overlapped Chinese handwriting recognition method aimed at the mobile platform. The multi-modal error correction method uses the voice speech modality to correct the handwriting recognition errors. At the same time, the N-best candidate list is also combined with the multi-modal method for correcting errors, and the artificial neural networks are used for sorting these candidates.In this thesis, we also proposed an overlapped Chinese handwriting recogniton method, which makes it possible for the users to write continuously on the same input area of the mobile device. This recognition method first extracts feature vectors from the original input strokes, and then does classification using the support vector machine or aritificial neural network. Moreover, this thesis also proposed a conversion method for turning the floating point calculation into the fixed point calculation, which largely enhancing the computing efficiency on mobile devices. The experiment results show that the multi-modal error correction method could rectify the handwriting recognition error efficiently, the overlapped Chinese handwriting recognition method could obtain high recognition rate and fully meet the recogniton requirement. All these algorithms can run stablely and efficiently on the mobile devices. |
PDF Full Text Request