Noisy-Aware Unsupervised Domain Adaptation for Scene Text Recognition

被引:0
作者
Liu, Xiao-Qian [1 ]
Zhang, Peng-Fei [2 ]
Luo, Xin [1 ]
Huang, Zi [2 ]
Xu, Xin-Shun [1 ]
机构
[1] Shandong Univ, Sch Software, Jinan 250101, Peoples R China
[2] Univ Queensland, Sch Elect Engn & Comp Sci, Brisbane, Qld 4072, Australia
基金
中国国家自然科学基金;
关键词
Text recognition; domain adaptation; entropy; noisy-aware; consistency regularization; NETWORK;
D O I
10.1109/TIP.2024.3492705
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Unsupervised Domain Adaptation (UDA) has shown promise in Scene Text Recognition (STR) by facilitating knowledge transfer from labeled synthetic text (source) to more challenging unlabeled real scene text (target). However, existing UDA-based STR methods fully rely on the pseudo-labels of target samples, which ignores the impact of domain gaps (inter-domain noise) and various natural environments (intra-domain noise), resulting in poor pseudo-label quality. In this paper, we propose a novel noisy-aware unsupervised domain adaptation framework tailored for STR, which aims to enhance model robustness against both inter- and intra-domain noise, thereby providing more precise pseudo-labels for target samples. Concretely, we propose a reweighting target pseudo-labels by estimating the entropy of refined probability distributions, which mitigates the impact of domain gaps on pseudo-labels. Additionally, a decoupled triple-P-N consistency matching module is proposed, which leverages data augmentation to increase data diversity, enhancing model robustness in diverse natural environments. Within this module, we design a low-confidence-based character negative learning, which is decoupled from high-confidence-based positive learning, thus improving sample utilization under scarce target samples. Furthermore, we extend our framework to the more challenging Source-Free UDA (SFUDA) setting, where only a pre-trained source model is available for adaptation, with no access to source data. Experimental results on benchmark datasets demonstrate the effectiveness of our framework. Under the SFUDA setting, our method exhibits faster convergence and superior performance with less training data than previous UDA-based STR methods. Our method surpasses representative STR methods, establishing new state-of-the-art results across multiple datasets.
引用
收藏
页码:6550 / 6563
页数:14
相关论文
共 50 条
[41]   Adversarial Entropy Optimization for Unsupervised Domain Adaptation [J].
Ma, Ao ;
Li, Jingjing ;
Lu, Ke ;
Zhu, Lei ;
Shen, Heng Tao .
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS, 2022, 33 (11) :6263-6274
[42]   Conditional Independence Induced Unsupervised Domain Adaptation [J].
Xu, Xiao-Lin ;
Xu, Geng-Xin ;
Ren, Chuan-Xian ;
Dai, Dao-Qing ;
Yan, Hong .
PATTERN RECOGNITION, 2023, 143
[43]   Unsupervised Domain Adaptation via Importance Sampling [J].
Xu, Xuemiao ;
He, Hai ;
Zhang, Huaidong ;
Xu, Yangyang ;
He, Shengfeng .
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, 2020, 30 (12) :4688-4699
[44]   Task Nuisance Filtration for Unsupervised Domain Adaptation [J].
Uliel, David ;
Giryes, Raja .
IEEE OPEN JOURNAL OF SIGNAL PROCESSING, 2025, 6 :303-311
[45]   Rethinking text rectification for scene text recognition [J].
Ke, Wenjun ;
Wei, Jianguo ;
Hou, Qingzhi ;
Feng, Hui .
EXPERT SYSTEMS WITH APPLICATIONS, 2023, 219
[46]   Class-Aware Mask-guided feature refinement for scene text recognition [J].
Yang, Mingkun ;
Yang, Biao ;
Liao, Minghui ;
Zhu, Yingying ;
Bai, Xiang .
PATTERN RECOGNITION, 2024, 149
[47]   Attention Guidance by Cross-Domain Supervision Signals for Scene Text Recognition [J].
Xue, Fanfu ;
Sun, Jiande ;
Xue, Yaqi ;
Wu, Qiang ;
Zhu, Lei ;
Chang, Xiaojun ;
Cheung, Sen-Ching .
IEEE TRANSACTIONS ON IMAGE PROCESSING, 2025, 34 :717-728
[48]   Part-aware Progressive Unsupervised Domain Adaptation for Person Re-Identification [J].
Yang, Fan ;
Yan, Ke ;
Lu, Shijian ;
Jia, Huizhu ;
Xie, Don ;
Yu, Zongqiao ;
Guo, Xiaowei ;
Huang, Feiyue ;
Gao, Wen .
IEEE TRANSACTIONS ON MULTIMEDIA, 2021, 23 :1681-1695
[49]   Semantic-Aware Contrastive Adaptation Bridges Domain Discrepancy for Unsupervised Remote Sensing [J].
Zhang, Liwen ;
Xu, Ting ;
Zeng, Changyang ;
Hao, Qi ;
Chen, Zhenghan ;
Liang, Xiaoxuan .
IEEE ACCESS, 2024, 12 :140734-140747
[50]   A Semantic Segmentation Method of Traffic Scene Based on Categories-Aware Domain Adaptation [J].
Jia Y. ;
Lang C. ;
Feng S. .
Jisuanji Yanjiu yu Fazhan/Computer Research and Development, 2020, 57 (04) :876-887