Robust End-to-End Offline Chinese Handwriting Text Page Spotter with Text Kernel

Offline Chinese handwriting text recognition is a long-standing research topic in the field of pattern recognition. In previous studies, detection and are separated, which leads to fact that highly dependent on results. this paper, we propose robust end-to-end page spotter framework. It unifies with kernel integrates global feature information optimize from multiple scales, reduces dependence improves robustness system. Our method achieves state-of-the-art results CASIA-HWDB2.0-2.2 dataset ICDAR-2013 competition dataset. Without any language model, correct rates 99.12% 94.27% for line-level recognition, 99.03% 94.20% page-level respectively. Code will be available at GitHub.