Research on Arbitrary Shape Tibetan Text Detection with Graph Network

Zezhou Xu, Jie Zhu, Yashan Liu, Zehui Xu, Songsi Yan, Chao Wang · 2022 International Conference on Image Processing, Computer Vision and Machine Learning (ICICML) · 2022

Text detection is an important field in computer vision. The research on Chinese and English text detection has achieved excellent results. But Tibetan text detection still needs further development. We use a relational reasoning graph network to improve the detection of arbitrary-shaped Tibetan text. The method is divided into two parts, text proposal part based on Convolutional Neural Network (CNN), and relational reasoning based on Graph Convolutional Network (GCN). Experiments on Tibetan text image datasets demonstrate the excellent performance.

Read the paper · More papers on PaperTik