ABSTRACT
Chinese character font recognition is a vital research area with applications in text recognition, artistic font design, and handwriting recognition. Traditional methods for this task are time-consuming, labor-intensive, and error-prone. The rise of deep learning has opened new possibilities in this field. However, neural network models often struggle with generalization and robustness, especially with new data, leading to catastrophic forgetting of old data. To address this, we propose an optimization algorithm for the Chinese character font recognition model SwordNet, combining parameter isolation and regularization. We designed an incremental module for SwordNet, enhancing it for better performance and scalability. The loss function incorporates a regularization term based on knowledge distillation principles, limiting significant parameter changes for old data and ensuring the model retains knowledge while learning new data. To tackle the diversity of Chinese character fonts and aid model recognition, we introduced a data augmentation method using noise, rotation, and occlusion. Experiments demonstrate that our regularized incremental learning optimization algorithm achieves a 98.81% accuracy rate with only new data, comparable to models trained on both new and old datasets. Notably, it also reduces training time by approximately 30%, marking a significant improvement in efficiency and accuracy for Chinese character font recognition.
- Abhishek Aich. 2021. Elastic Weight Consolidation (EWC): Nuts and Bolts. arxiv:2105.04093Google Scholar
- Rahaf Aljundi, Lucas Caccia, Eugene Belilovsky, Massimo Caccia, Min Lin, Laurent Charlin, and Tinne Tuytelaars. 2019. Online Continual Learning with Maximally Interfered Retrieval. arxiv:1908.04742Google Scholar
- Xiang Bai, Baoguang Shi, Chengquan Zhang, Xuan Cai, and Li Qi. 2017. Text/Non-Text Image Classification in the Wild with Convolutional Neural Networks. Pattern Recogn. 66, C (jun 2017), 437–446. https://doi.org/10.1016/j.patcog.2016.12.005Google ScholarDigital Library
- Francisco M. Castro, Manuel J. Marín-Jiménez, Nicolás Guil, Cordelia Schmid, and Karteek Alahari. 2018. End-to-End Incremental Learning. arxiv:1807.09536 [cs.CV]Google Scholar
- Arslan Chaudhry, Marc’Aurelio Ranzato, Marcus Rohrbach, and Mohamed Elhoseiny. 2019. Efficient Lifelong Learning with A-GEM. arxiv:1812.00420Google Scholar
- Jingchao Chen, Shiyi Mu, Shugong Xu, and Youdong Ding. 2021. HENet: Forcing a Network to Think More for Font Recognition. arxiv:2110.10872 [cs.CV]Google Scholar
- Matthias Delange, Rahaf Aljundi, Marc Masana, Sarah Parisot, Xu Jia, Ales Leonardis, Greg Slabaugh, and Tinne Tuytelaars. 2021. A continual learning survey: Defying forgetting in classification tasks. IEEE Transactions on Pattern Analysis and Machine Intelligence (2021), 1–1. https://doi.org/10.1109/tpami.2021.3057446Google ScholarCross Ref
- Arthur Douillard, Matthieu Cord, Charles Ollion, Thomas Robert, and Eduardo Valle. 2020. PODNet: Pooled Outputs Distillation for Small-Tasks Incremental Learning. arxiv:2004.13513 [cs.CV]Google Scholar
- Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. 2015. Distilling the Knowledge in a Neural Network. arxiv:1503.02531 [stat.ML]Google Scholar
- Xinting Hu, Kaihua Tang, Chunyan Miao, Xian-Sheng Hua, and Hanwang Zhang. 2021. Distilling Causal Effect of Data in Class-Incremental Learning. arxiv:2103.01737 [cs.AI]Google Scholar
- James Kirkpatrick, Razvan Pascanu, Neil Rabinowitz, Joel Veness, Guillaume Desjardins, Andrei A. Rusu, Kieran Milan, John Quan, Tiago Ramalho, Agnieszka Grabska-Barwinska, Demis Hassabis, Claudia Clopath, Dharshan Kumaran, and Raia Hadsell. 2017. Overcoming catastrophic forgetting in neural networks. Proceedings of the National Academy of Sciences 114, 13 (mar 2017), 3521–3526. https://doi.org/10.1073/pnas.1611835114Google ScholarCross Ref
- Xudong Li, Jingyi Wang, Haiyang Zhang, Yongke Huang, and Huihui Huang. 2022. SwordNet: Chinese Character Font Style Recognition Network. IEEE Access 10 (2022), 8388–8398. https://doi.org/10.1109/ACCESS.2022.3143795Google ScholarCross Ref
- Zhizhong Li and Derek Hoiem. 2017. Learning without Forgetting. arxiv:1606.09282Google Scholar
- Charles X. Ling and Tanner Bohn. 2021. A Deep Learning Framework for Lifelong Machine Learning. arxiv:2105.00157Google Scholar
- Zheda Mai, Ruiwen Li, Jihwan Jeong, David Quispe, Hyunwoo Kim, and Scott Sanner. 2021. Online Continual Learning in Image Classification: An Empirical Survey. arxiv:2101.10423Google Scholar
- Arun Mallya and Svetlana Lazebnik. 2018. PackNet: Adding Multiple Tasks to a Single Network by Iterative Pruning. arxiv:1711.05769 [cs.CV]Google Scholar
- Sylvestre-Alvise Rebuffi, Alexander Kolesnikov, Georg Sperl, and Christoph H. Lampert. 2017. iCaRL: Incremental Classifier and Representation Learning. arxiv:1611.07725Google Scholar
- Joan Serrà, Dídac Surís, Marius Miron, and Alexandros Karatzoglou. 2018. Overcoming catastrophic forgetting with hard attention to the task. arxiv:1801.01423Google Scholar
- Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou. 2020. MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices. arxiv:2004.02984Google Scholar
- Shipeng Yan, Jiangwei Xie, and Xuming He. 2021. DER: Dynamically Expandable Representation for Class Incremental Learning. arxiv:2103.16788 [cs.CV]Google Scholar
Index Terms
- Optimization of Chinese Character Font Recognition Model Based on Incremental Learning
Recommendations
Semantic-Based Handwritten Chinese Character Recognition Model
ICCMS '10: Proceedings of the 2010 Second International Conference on Computer Modeling and Simulation - Volume 01There have been many different literals discussing algorithms for handwritten Chinese character recognition, but most algorithms aim at recognizing isolated Chinese character one by one. Therefore, their recognition accuracy isn’t good enough for the ...
Radical aggregation network for few-shot offline handwritten Chinese character recognition
Highlights- A radical aggregation network for radical-level Chinese character recognition.
- ...
Graphical abstractOverall architecture of our radical aggregation network (RAN). The RAN consists of three components, a radical mapping encoder (RME), a radical aggregation module (RAM), and a character analysis decoder (CAD). The RME module ...
AbstractOffline handwritten Chinese character recognition has attracted much interest due to its various applications. The most cutting-edge methods treat Chinese character as a whole, ignoring the structures and radicals that compose ...
LSH-based large scale chinese calligraphic character recognition
JCDL '13: Proceedings of the 13th ACM/IEEE-CS joint conference on Digital librariesChinese calligraphy is the art of handwriting and is an important part of Chinese traditional culture. But due to the complexity of shape and styles of calligraphic characters, it is difficult for com-mon people to recognize them. So it would be great ...
Comments