SAN: Structure-Aware Network for Complex and Long-tailed Chinese Text Recognition

TL;DR

This paper introduces SAN, a structure-aware network that leverages hierarchical composition information to improve recognition of complex and tail Chinese characters, addressing shape confusion and data imbalance issues.

Contribution

The paper proposes a novel structure-aware network with an auxiliary radical branch and a Tree-Similarity-based weighting mechanism for better Chinese text recognition.

Findings

Significant improvement in recognizing complex characters.

Enhanced performance on tail classes.

Effective utilization of hierarchical composition information.

Abstract

In text recognition, complex glyphs and tail classes have always been factors affecting model performance. Specifically for Chinese text recognition, the lack of shape-awareness can lead to confusion among close complex characters. Since such characters are often tail classes that appear less frequently in the training-set, making it harder for the model to capture its shape information. Hence in this work, we propose a structure-aware network utilizing the hierarchical composition information to improve the recognition performance of complex characters. Implementation-wise, we first propose an auxiliary radical branch and integrate it into the base recognition network as a regularization term, which distills hierarchical composition information into the feature extractor. A Tree-Similarity-based weighting mechanism is then proposed to further utilize the depth information in the hierarchical representation. Experiments demonstrate that the proposed approach can significantly improve the performances of complex characters and tail characters, yielding a better overall performance. Code is available at https://github.com/Levi-ZJY/SAN.