Online gesture-based interaction with visual oriental characters based on manifold learning

Online gesture-based interaction with characters has become a more natural and informative human-computer interface with the popularity of new interactive devices (e.g., Kinect and Leap Motion). In this paper, a new feature descriptor named Segmented Directed-edge Vector (SDV) is proposed. This simple and yet quite effective descriptor is able to capture the characteristics of visual oriental characters. Moreover, we explicitly build the mappings from SDVs to features in a subspace by a modified Locality Preserving Projections (LPP) method with stroke class constraints. These mappings can yield meaningful subspace structures for larger character sets. Extensive experiments on the online interactive system demonstrate the robustness of our method to various issues in gesture-based characters input, such as unnatural breaks, overlapped or distorted radicals, and unconscious or quivering trajectories. Our system can still achieve accurate recognition when accumulative errors occur with complex characters.