Open AccessAdvanced Chinese Character Detection for Natural Scene Based on EAST
Author(s) -
Jianxin Zhang,
Yu Feng
Publication year - 2020
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1550/3/032050
Subject(s) - character (mathematics) , artificial intelligence , computer science , text detection , pattern recognition (psychology) , feature (linguistics) , feature extraction , task (project management) , field (mathematics) , image (mathematics) , natural (archaeology) , character recognition , computer vision , mathematics , geography , linguistics , engineering , philosophy , geometry , systems engineering , archaeology , pure mathematics
Currently in the field of image vision processing, text detection and text recognition under natural scenes is a challenging task. At present, most of the research on text positioning is mainly aimed at English, and the detection and recognition of Chinese characters is relatively few. EAST text detection algorithm is simple in structure and high in accuracy. However, because of the feature extraction network, there is a small sense field, which cannot be well adapted to the text detection of Chinese characters. In order to adapt it to the natural scene Chinese character text detection, the feature extraction network of the modified EAST algorithm is Mobilnet-V2 and Resnet50, which increases the depth of the network and extracts more image features. The modified algorithm is tested on the dataset MSRA-TD500, and the obtained accuracy shows that the improved network performance is better than the original algorithm.