Research On Recognition Of Online Handwritten Mathematical Expressions

With the rapid development of network technology and computer technology,electronic documents become an integral part of daily study and work. As mathematicalexpression has been an essential part of many electronic documents, how totypeit intothe computer is a serious problem. Originally, people mainly input mathematicalexpression through graphical interface editor and one-dimensional description language.However, graphical editing tools existsshortcomings,such as poor efficiency andcumbersome operations. Besides, it is necessary to master the language before typingthe description language for mathematical expressions. With the development andpopularization of handwriting technology and equipments, more intuitive andconvenient on-line recognition of handwritten mathematical expression has becomeanother important way to input mathematical expressions.Generally, online handwritten mathematical expression recognition process isdivided intothree consecutive steps, including symbol segmentation, characterrecognition and structural analysis. Existing methods have the same identificationprocess, but the specific methods usingin each step are different. When entering a newstroke, previous recognition results are often discarded incurrent identificationmethods.Then the whole recognition process will be restarted without supportingwrite-back strokes. Obviously, the efficiency and capacity of traditional methods are indemand of improvement. This paper aims atproposing an optimal identification methodby studying the impact of new input stroke on the region and the adaptive ability oforganizational structure of the expression, especially for the dynamic modification ofstrokes.The tasks are displayed as follow:1) This paperanalyzes a large number of handwriting mathematical expressioninput instance.It is found that newly input stroke in most cases affect only a part of theprevious input mathematical expressions. Therefore, a real-time identification methodbased on the positioning of the stroke’s affected area is proposed. The method bases onprevious recognition resultsand stores the previous recognition result in themathematical expression tree. Then, search the new stroke’s affected area inthemathematical expression tree. Finally, complete the expression recognition byupdating the mathematical expression tree according to the relationshipbetween thenodes in affected area and the new stroke. 2)Because of the expansive search space of symbol segmentation, the existingmethods have not solved the write-back problem of stroke. In this paper, a strokeaffecting regional positioning algorithm is proposed, reducing the size of collection ofstroke in the identification process into manageable size, thus solving the problem ofstroke write-back.3) On the basis of these studies, an online handwriting mathematical expressionrecognition prototype system is designed and implemented.The prototype system istested on standard test set and compared with other systems.The results showed that therecognition speed and recognition rate of the proposed method are both improved.