CRCRNet༚An Integrated Approach to Table Data Extraction
S.-W. Sun, Jiewen Zhu, Chunlei Zhou, Chenchen Yang · 2024
This paper presents an algorithm for extracting tabular data. Among them, the text structure recognition uses ResNet as the backbone network, and the FPN module and TS-CB module are added to increase the detection performance of the model for edge distribution. Method of text prediction network to optimize the structure of CRNN, the CRCRNet model was proposed, and the row reconstruction module and column reconstruction module were used in the network to improve the recognition and prediction ability of the model. The process of text recognition prediction and table reconstruction are integrated into a whole, and a complete table recognition algorithm is proposed. In Chapter 5, the experimental results on ICDAR2013 data set are given and compared with other form recognition methods. The experimental results show that the proposed recognition algorithm has better performance.