Research On Convolutional Neural Network Based Handwritten Formula Symbol Recognition Method

With the widespread use of mobile terminal devices such as smart phones and tablet computers,a large number of handwritten electronic documents have emerged,which handwriting technology plays an important role.Handwritten formula symbol recognition is an important part of handwriting recognition research,which promotes the development of handwriting field to a certain extent.However,handwritten formula symbol recognition is a very challenging problem.On the one hand,there are a large number of handwriting styles of the same symbol;on the other hand,there are a large number of similar symbols,which leads to confusion during the recognition stages.Therefore,regarding the characteristics of handwritten formula symbols,this paper proposes two research methods based on convolutional neural network recognition.The main tasks as follows:1 、 This paper proposes an algorithm for offline handwriting formulas symbol recognition based on dense convolutional neural network.we use data-driven methods instead of traditional methods,and its characteristics do not require human intervention to obtain features.Meanwhile,the network has a dense residual structure,which effectively reduces the disappearance of the gradient and extracts deep-level features of the network.Fine-grained features of the shallow layer are ingeniously integrated with the depth features in a self-connecting manner.The standard mathematical formula symbol library provided by the Competition Organization on Recognition of Online Handwritten Mathematical Expression(CROHME)is used to verify the proposed algorithm.Experimental results show that the proposed algorithm can effectively learn the attributes of handwritten formula symbol recognition,which is better than the existing algorithms for handwritten formula symbol recognition.2、 Regarding to the lack of a dynamic trajectory information during the process of converting online data to offline images,it is difficult to distinguish similarity symbols.Therefore,in order to solve the above problems,this paper proposes a multifeature learning handwritten mathematical formula symbol recognition algorithm through joint training.This algorithm extracts online eight-direction feature maps from the original sequence.To make up for the lack of structural information of directional features,a Gabor filter is used to decompose the image in directions to obtain a multidirectional gradient map.Secondly,the directional feature layer is integrated into the convolutional layer to learn online and offline features respectively.Finally,the construction of joint losses improves the discriminative features by means of selfsupervised learning.Experimental results show that the proposed algorithm significantly improves the discrimination of similar symbols and thus improves the accuracy of recognition.