An Intelligent Chinese Handwriting Tool With Stroke Error Detection

Due to the complex shapes and various writing styles of Chinese characters, the students might make some potential errors in their handwritings which include stroke production, sequence and relationship errors. With the Chinese handwriting education tool, it can automatically find the errors in the students'handwriting and provide feedback which is helpful to improve their writing skill. The main research works and contributions are described as follows.The attributed relational graph (ARG) is used to represent both the template and sample Chinese character. The nodes in the ARG are used to describe the strokes of the character and the edges in the ARG denote the spatial relations between any two strokes. Due to the complex structure of the Chinese character, we are motivated to generate a refined spatial relationship to illustrate the structural relationship between strokes in a character. And K-means clustering is applied to derive our refined spatial relationship. The refined spatial relationship increases its granularity by considering the relative distance so that it contains more structural information about the Chinese character. At the same time, with the proposed refined spatial relationship, we can detect the stroke relationship error which has seldom been mentioned in the previous work.During the matching process between the template and sample Chinese character, the relationship between strokes are different if we apply some deformation (Translating, scaling et al.) on the one of the stroke. And those differences can be represented by the interval neighborhood graph. We generate a refined interval neighborhood graph by considering the deformations among our refined spatial relationships. The edges between two neighbors that describe changes between them are associated with uniform weights. To account for the fact that changes in various interval relationships may have different significance, we propose two ways to adapt the weights of the interval neighborhood graph according to the significance of change in spatial relationships observed in the training data. The first method is based on statistical analysis of relationship change. Since this method need manually checking of the significant changing, then we propose the second method relies on back-propagation neural network to automatically generate the weight-adaptive interval neighborhood graph. Due to the noise and distortion of the students'handwritings, the error-tolerant graph matching (etgm) with corresponding node and edge edit operations is applied to find the optimal matching between template and sample character. With the proposed matching, we can locate the stroke production and stroke order errors with low error rate (reduce 30% at most,10% at least) compared with the previous work. Based on the case analysis of the handwriting errors in students' handwritings, our proposed Chinese handwriting education system can detect more types of handwriting error than the existing works.