融合多层语义特征图的缅甸语图像文本识别方法(Burmese Image Text Recognition Method Fused with Multi-layer Semantic Feature Maps)

Fuhao Liu (刘福浩), Cunli Mao (毛存礼), Zhengtao Yu (余正涛), Chengxiang Gao (高盛祥), Linqin Wang (王琳钦), Xuyang Xie (谢旭阳)


Abstract
由于缅甸语存在特殊的字符组合结构,在图像文本识别研究方面存在较大的困难,直接利用现有的图像文本识别方法识别缅甸语图片存在字符缺失和复杂背景下识别效果不佳的问题。因此,本文提出一种融合多层语义特征图的缅甸语图像文本识别方法,利用深度卷积网络获得多层图像特征并对其融合获取多层语义信息,缓解缅甸语图像中由于字符嵌套导致特征丢失的问题。另外,在训练阶段采用MIX UP的策略进行网络参数优化,提高模型的泛化能力,降低模型在测试阶段对训练样本产生的依赖。实验结果表明,提出方法相比基线模型准确率提升了2.2%。
Anthology ID:
2021.ccl-1.17
Volume:
Proceedings of the 20th Chinese National Conference on Computational Linguistics
Month:
August
Year:
2021
Address:
Huhhot, China
Venue:
CCL
SIG:
Publisher:
Chinese Information Processing Society of China
Note:
Pages:
176–185
Language:
Chinese
URL:
https://aclanthology.org/2021.ccl-1.17
DOI:
Bibkey:
Cite (ACL):
Fuhao Liu, Cunli Mao, Zhengtao Yu, Chengxiang Gao, Linqin Wang, and Xuyang Xie. 2021. 融合多层语义特征图的缅甸语图像文本识别方法(Burmese Image Text Recognition Method Fused with Multi-layer Semantic Feature Maps). In Proceedings of the 20th Chinese National Conference on Computational Linguistics, pages 176–185, Huhhot, China. Chinese Information Processing Society of China.
Cite (Informal):
融合多层语义特征图的缅甸语图像文本识别方法(Burmese Image Text Recognition Method Fused with Multi-layer Semantic Feature Maps) (Liu et al., CCL 2021)
Copy Citation:
PDF:
https://aclanthology.org/2021.ccl-1.17.pdf