Gestures And Two Classifier-based On-line Handwritten Formula Recognition

Recognition-based pen user interface is one of the most important post-WIMP interfaces. Symbols including data symbols and command symbols are composed of strokes, which is the main data type of pen interface environment. The base operation on strokes is recognition, so the construction of pen applications needs a high recognition rate. However, because of ambiguity of pen interfaces, current pattern recognition technologies can't resolve all problems. Many interaction technologies are necessary.On-line handwritten mathematics recognition is an important area of pen user interface application, which is widely used in hand mobile computing devices, tablet pc and electronic whiteboard. This project represents the problems faced by pen user interface applications. While the early application systems are all localized in the frame of pattern recognition, it is inconvenient to use, and recognition rate should be improved further. In this paper, we stick to the user-centered, natural and implicit design thinking, face the specific application of on-line handwritten mathematics recognition, aim at the improvement of previous systems, make advances in data symbols recognition, gesture command interaction and interaction technology research.In the research on data symbol recognition, we design a two-stage multi-features classifier with threshold, which improves recognition rate, and at same time avoid adding much computing time. I believe that this method is also applicable to other pen user interface applications.In the research on gesture interaction, we use it as the main command interaction method under the pen user interface environment. We work on three sides: design, recognition and evaluation, and implement a gesture set for on-line handwritten mathematics recognition.In the research on interaction technologies, we focus on context-aware interaction, which is applied on mathematics recognition successfully. We also apply technologies such as best choice, direct assignment, local magnification, and supplementing strokes to resolve problems that can't be settled by pattern recognition alone.