Context-Aware Confidence Estimation for Rejection in Handwritten Chinese Text Recognition

被引:0
作者
Liu, Yangyang [1 ,2 ]
Chen, Yi [1 ,2 ]
Yin, Fei [1 ,2 ]
Liu, Cheng-Lin [1 ,2 ]
机构
[1] Chinese Acad Sci, Inst Automat, State Key Lab Multimodal Artificial Intelligence, Beijing 100190, Peoples R China
[2] Univ Chinese Acad Sci, Sch Artificial Intelligence, Beijing 100049, Peoples R China
来源
DOCUMENT ANALYSIS AND RECOGNITION-ICDAR 2024, PT I | 2024年 / 14804卷
基金
中国国家自然科学基金;
关键词
Handwritten Chinese Text Recognition; Confidence Estimation; Geometric Context; Bayesian probability formula; DISCRIMINATIVE UTTERANCE VERIFICATION; TRANSFORMATION; ONLINE;
D O I
10.1007/978-3-031-70533-5_9
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Handwritten Chinese Text Recognition (HCTR) has been advanced largely by deep learning in recent years. However, the remaining recognition errors still hinder reliability-critical applications where zero-error is desired. Rejecting low-confidence patterns can help reduce the error rate but the increased rejection rate is also harmful. In this paper, we propose a character confidence estimation method incorporating contexts for character rejection in HCTR. Based on a text line recognizer outputting character segmentation and classification results, the confidence of each segmented character is estimated by combining the scores of a re-trained character classifier, the linguistic and geometric contexts. We introduce a probabilistic formula for estimating the confidence by combining the classifier and contextual scores, and an improved approach for scoring the geometric context using unary and binary geometric features. Experimental evaluations on the CASIA-HWDB and ICDAR2013 datasets demonstrate that our method can significantly improve the rejection performance in respect of low error rate at moderate rejection rate. The re-trained classifier, the linguistic context and the geometric context are all justified effective to improve the confidence.
引用
收藏
页码:134 / 151
页数:18
相关论文
共 33 条
[31]   PageNet: Towards End-to-End Weakly Supervised Page-Level Handwritten Chinese Text Recognition [J].
Dezhi Peng ;
Lianwen Jin ;
Yuliang Liu ;
Canjie Luo ;
Songxuan Lai .
International Journal of Computer Vision, 2022, 130 :2623-2645
[32]   Minimum Risk Training for Handwritten Chinese/Japanese Text Recognition Using Semi-Markov Conditional Random Fields [J].
Zhou, Xiang-Dong ;
Tian, Feng ;
Liu, Cheng-Lin ;
Wang, Hong-An .
2013 12TH INTERNATIONAL CONFERENCE ON DOCUMENT ANALYSIS AND RECOGNITION (ICDAR), 2013, :940-944
[33]   MQDF WITH A NOVEL COVARIANCE MATRIX ESTIMATION AND DISCRIMINANT LSRC, WHICH IS BETTER FOR IN-AIR HANDWRITTEN CHINESE CHARACTER RECOGNITION [J].
Xu, Zhaochun ;
Wang, Weiqiang ;
Lu, Ke .
2016 IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING (ICIP), 2016, :3274-3278