Research And Realization On Handwriting Recognition Of The Online Mathematical Formula

In the education industry, in order to track students' learning trajectory and weak knowledge in real-time, and then provide students with accurate teaching assistant services . Thus, The main research content of this paper is online mathematical formula handwriting recognition. It aims to propose a robust and feasible solution to identify students' handwritten mathematical formula.The content of this research can be summarized as follows:Firstly, proposed a single character recognition model combining CNN and DBN .There are a lot of ambiguities in handwritten mathematical formulas. At the same time there are considerable difficulties in judging accurately the two-dimensional structure, besides, many similar characters in mathematical formulas that are easily confused. These conditions increase the difficulty of automatic identification of the machine. In this paper, a combinatorial recognition algorithm based on combinatorial ordering is proposed. In this algorithm, these uncertain cases are saved as candidates to generate the combined paths, then sorts these paths based on the phrase frequency table,semantic model, and recognition confidence , such that, this algorithm can guarantee the recognition correctness and greatly simplify the complexity of the system as well as increasing the recognition robustness, at the same time.Secondly, proposed a combining and ordering based algorithm for handwriting recognition.A single-character classifier is built and trained by Convolutional Neural Network(CNN). In view of the vulnerability of deep neural network to the anti-samples, this paper proposes a deep belief system (Deep Belief Network) , DBN) to encode the data for each category, using it as a recognition confidence determination model that can maintain a small error for the same sample decoding reconstruction, combining this confidence rating with the confidence evaluation given by CNN improves the ability to reject the sample.Thirdly, proposed a quick learning method based on the wrong case.In this paper, we propose a learning method based on the wrong case, we can quickly learn a new combination mapping knowledge which will be regarded an combination knowledge to add to the system , to avoid the same recognition mistakes again.Based on the above method, a recognition system is implemented. The experimental results show that the online mathematical formula handwriting recognition method proposed in this paper has high recognition rate and good robustness...