Alert button

Maximum Entropy Regularization and Chinese Text Recognition

Jul 09, 2020
Changxu Cheng, Wuheng Xu, Xiang Bai, Bin Feng, Wenyu Liu

Figure 1 for Maximum Entropy Regularization and Chinese Text Recognition
Figure 2 for Maximum Entropy Regularization and Chinese Text Recognition
Figure 3 for Maximum Entropy Regularization and Chinese Text Recognition
Figure 4 for Maximum Entropy Regularization and Chinese Text Recognition

Share this with someone who'll enjoy it:

Chinese text recognition is more challenging than Latin text due to the large amount of fine-grained Chinese characters and the great imbalance over classes, which causes a serious overfitting problem. We propose to apply Maximum Entropy Regularization to regularize the training process, which is to simply add a negative entropy term to the canonical cross-entropy loss without any additional parameters and modification of a model. We theoretically give the convergence probability distribution and analyze how the regularization influence the learning process. Experiments on Chinese character recognition, Chinese text line recognition and fine-grained image classification achieve consistent improvement, proving that the regularization is beneficial to generalization and robustness of a recognition model.

* 15 pages  
View paper onarxiv icon

Share this with someone who'll enjoy it: