End-To-End System Recognition Based on Improved Faster RCNN+CTC Text Detection
Download as PDF
DOI: 10.23977/ESAC2020024
Author(s)
Ying Wang, Baolong Guo, Zhe Huang and Cheng Li
Corresponding Author
Baolong Guo
ABSTRACT
In the current era of rapid development of science and technology information, and the rapid development of Internet technology and mobile terminal equipment, images have gradually become the main source of people's daily communication. The technology of extracting content in the image has become a focus of attention. This article mainly uses improved Faster R-CNN text detection and recognition end-to-end system to extract and recognize text information in images. This algorithm introduces a bidirectional LSTM network in the original Faster R-CNN network to retain the context information of the text, and uses Monte Carlo non-maximum suppression method to judge the slanted text box. This method detects the performance of long text and slanted angle text. There has been a significant improvement. Finally, CTC is used for text recognition.
KEYWORDS
Text Detection; Text Recognition; Faster R-CNN; LSTM; CTC