蒙古文的一大特点是字符无缝连接,因此一个蒙古文单词有多种字符划分方式。根据蒙古文这一特点,该文提出了多尺度蒙古文脱机手写识别方法,即让一个手写蒙古文单词图像对应多种目标序列,用多个目标序列同时约束训练模型,使得模型更加精准地学习手写图像的细节信息和蒙古文构词规则。该文提出了“十二字头”码、变形显现码和字素码3种字符划分方法,且拥有相互包含关系,即“十二字头”码可以分解为变形显现码、变形显现码可以进一步分解为字素码。多尺度模型首先用多层双向长短时记忆网络对序列化手写图像进行处理,之后加入第一层连接时序分类器做“十二字头”码序列的映射,然后是第二层连接时序分类器做变形显现码序列的映射,最后是第三层连接时序分类器做字素码序列的映射。用三个连接时序分类器损失函数的和作为模型的总损失函数。实验结果表明,该模型在公开的蒙古文脱机手写数据集MHW上表现出了最佳性能,在简单的最佳路径解码方式下,测试集Ⅰ上的单词识别准确率为66.22%、测试集Ⅱ上为63.97%。
Abstract
One major feature of Mongolian is the seamless connection of characters in a word, so a Mongolian word has multiple character division methods. A multi-scale Mongolian offline handwriting recognition method is proposed, in which one image of handwritten Mongolian word are mapped into to multiple target sequences to train the model. This paper distinguishes three candidate character division methods: "Twelve Prefix" code, presentation form code and grapheme code. The multi-scale model processes the sequence of handwritten images with a Bidirectional Long Short-Term Memory network, which are then fed into a Connectionist Temporal Classification (CTC) layer to map the image to the "Twelve Prefix" code sequence, the presentation form code sequence, and the grapheme code sequence, respectively. The sum of three CTC loss is used as the total loss function of the model. The experiments show that the model achieves the best performance on the public Mongolian offline handwritten data set MHW, with 66.22% and 63.97% accuracy on test set I and II, respectively.
关键词
蒙古文 /
脱机手写识别 /
多尺度 /
LSTM /
CTC
{{custom_keyword}} /
Key words
Mongolian /
offline handwriting recognition /
multi-scale /
LSTM /
CTC
{{custom_keyword}} /
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] Dutta K, Krishnan P, Mathew M, et al. Improving CNN-RNN hybrid networks for handwriting recognition[C]//Proceedings of International Conference on Frontiers in Handwriting Recognition. Niagara Falls, NY, USA:IEEE,2018:80-85.
[2] Ghosh R, Vamshi C, Kumar P. RNN based online handwritten word recognition in Devanagari and Bengali scripts using horizontal zoning[J]. Pattern Recognition, 2019, 92: 203-218.
[3] Granet A, Morin E, Mouchere H, et al. Separating optical and language models through encoder-decoder strategy for transferable handwriting recognition[C]//Proceedings of International Conference on Frontiers in Handwriting Recognition.Niagara Falls,Canada: IEEE,2018: 309-314.
[4] Graves A, Fernndez S, Gomez F, et al. Connectionist temporal classification: Labelling unsegmented sequence data with recurrent neural networks[C]//Proceedings of International Conference on Machine Learning. Pittsburgh, Pennsylvania, USA, 2006:369-376.
[5] Graves A,Schmidhuber J. Offline handwriting recognition with multidimensional recurrent neural networks[C]//Proceedings of Advances in Neural Information Processing Systems. London:Springer, 2009:545-552.
[6] Frinken V, Uchida S. Deep BLSTM neural networks for unconstrained continuous handwritten text recognition[C]//Proceedings of International Conference on Document Analysis and Recognition. Tunis, Tunisia:IEEE, 2015:911-915.
[7] Dutta K, Krishnan P, Mathew M, et al. Improving CNN-RNN hybrid networks for handwriting recognition[C]//Proceedings of International Conference on Frontiers in Handwriting Recognition, 2018:80-85.
[8] 刘聪.大词汇量脱机手写蒙古文整词识别研究[D].内蒙古: 内蒙古大学硕士学位论文,2019.
[9] 李进, 高静, 陈俊杰,等. 基于条件生成对抗网络的蒙古文字体风格迁移模型[J]. 中文信息学报, 2020, 34(04):58-62.
[10] Wang W, Wei H, Zhang H. End-to-end model based on bidirectional LSTM and CTC for segmentation-free traditional Mongolian recognition[C]//Proceedings of Chinese Control Conference. Guangzhou, China:IEEE,2019:8723-8727.
[11] Wei H, Liu C, Zhang H, et al. End-to-end model for offline handwritten Mongolian word recognition[C]//Proceedings of Natural Language Processing and Chinese Computing. Dunhuang, China:Springer,2019:220-230.
[12] 范道尔吉,高光来,武慧娟. MHW蒙古文脱机手写数据库及其应用[J]. 中文信息学报, 2018, 32(1): 89-95.
[13] 范道尔吉,高光来,武慧娟. 基于字素分割的蒙古文手写识别研究[J]. 中文信息学报, 2017, 31(5): 74-80.
[14] Fan D, Gao G. DNN-HMM for large vocabulary Mongolian offline handwriting recognition[C]//Proceedings of International Conference on Frontiers in Handwriting Recognition. IEEE, 2016:72-77.
[15] 确精扎布. 蒙古文编码[M]. 呼和浩特: 内蒙古大学出版社, 2000: 20-35.
[16] 中华人民共和国工业和信息化部. 信息技术蒙古文变形显现字符集和控制字符使用规则: GB/T 26226-2010[S]. 北京: 中国标准出版社,2011.
{{custom_fnGroup.title_cn}}
脚注
{{custom_fn.content}}
基金
国家自然科学基金(61763034);内蒙古自治区自然科学基金(2020MS06005)
{{custom_fund}}
PDF(1992 KB)
