A Symbol Dominance Based Formulae Recognition Approach for PDF Documents

Zhang, Xiaode; Gao, Liangcai; Yuan, Ke; Liu, Runtao; Jiang, Zhuoren; Tang, Zhi

doi:10.1109/icdar.2017.189

Cited by 13 publications

(8 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Automatic evaluation of the mathematical expression recognition is not an easy task due to the representation ambiguity of the ground truth data (usually as LaTeX or MathML) [2]. For example, x can be denoted as 'x{\prime}' or x in LaTeX.…”

Section: Evaluation Methodsmentioning

confidence: 99%

“…Survey papers [15] and [16] have given the detailed summarization. As a supplement, we roughly classified methods for mathematical expression recognition into three categories: rule based, grammar based, and deep learning based [2]. Previous researches most commonly process the symbol segmentation, symbol recognition, and structure analysis separately, and rule based methods are widely used.…”

Section: Related Workmentioning

confidence: 99%

“…Math symbols are commonly small objects and detailed information is very useful to identify them. (2) The image encoder should be easy to optimize and keep the capacity at the same time. The whole model, ConvMath, in order to achieve high performance, is relatively sophisticated and deep.…”

Section: A Image Encodermentioning

confidence: 99%

“…Mathematical expressions are indispensable to describe problems and theories in math, physics, and many other fields, thus play an important role in research and education [1]. Recognition of mathematical expressions aims at converting the two-dimensional (2-D) expression description into a onedimensional (1-D) structured sequence, which is significant for many downstream applications [2], such as scientific document digitization, mathematical information retrieval and reading to visually impaired users. Although some word processing tools support the input of mathematical expressions, it is timeconsuming or requires special skills [3].…”

Section: Introductionmentioning

confidence: 99%

“…However, due to some characteristics of mathematical expressions, recognition of them is a challenging task. For example, the spatial arrangement of math symbols exhibits complicated 2-D layout, including ABOVE, BELOW, LEFTUP, SUPERscript, SUBscript, CONTAIN, and so on, as mentioned in [2]. Different from OCR which most commonly recognizes 1-D linear characters one by one, mathematical expression recognition needs to recognize math symbols as well as the 2-D spatial relationships among them, and finally describe the expression using a markup language.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

ConvMath: A Convolutional Sequence Network for Mathematical Expression Recognition

Yan

Zhang

Gao

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

Despite the recent advances in optical character recognition (OCR), mathematical expressions still face a great challenge to recognize due to their two-dimensional graphical layout. In this paper, we propose a convolutional sequence modeling network, ConvMath, which converts the mathematical expression description in an image into a LaTeX sequence in an end-to-end way. The network combines an image encoder for feature extraction and a convolutional decoder for sequence generation. Compared with other Long Short Term Memory(LSTM) based encoder-decoder models, ConvMath is entirely based on convolution, thus it is easy to perform parallel computation. Besides, the network adopts multi-layer attention mechanism in the decoder, which allows the model to align output symbols with source feature vectors automatically, and alleviates the problem of lacking coverage while training the model. The performance of ConvMath is evaluated on an open dataset named IM2LATEX-100K, including 103556 samples. The experimental results demonstrate that the proposed network achieves stateof-the-art accuracy and much better efficiency than previous methods.

show abstract

Section: Evaluation Methodsmentioning

confidence: 99%