Searching from the Prediction of Visual and Language Model for Handwritten Chinese Text Recognition

Liu, Brian; Sun, Wei; Kang, Wei; Xu, Xianchao

doi:10.1007/978-3-030-86334-0_18

Cited by 10 publications

(10 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Specifically, the latest results of the CTC/attention-based approaches [14], [60], which were updated by the authors of [28] at their website 2 , are presented in Table III. The performances of the other two methods [65], [30] are obtained from our 2 https://github.com/HCIILAB/SCUT-HCCDoc Dataset Release reimplementation based on their official codes 3,4 . It can be observed that our method achieves state-of-the-art performance on SCUT-HCCDoc with an AR of 90.71% and a CR of 92.01%.…”

Section: ) Experimental Resultsmentioning

confidence: 99%

“…Xie et al [21] improved the CTC-based methods by exploring data augmentation and preprocessing. Liu et al [65] explored residual and squeeze-and-excitation structures for feature extraction and proposed context beam search to integrate the Transformerbased [63] language model into CTC-based methods. The attention mechanism, which has been widely adopted in scene text recognition [35], [30], [67], action recognition [37], [38], and video processing [39], can also be applied to offline HCTR.…”

Section: A Offline Handwritten Chinese Text Recognitionmentioning

confidence: 99%

Section: Transcriptionmentioning

confidence: 99%

“…Recently, the Transformer-based [63] language model has emerged in the field of HCTR [40], [64], [65]. After preparing CTC-style predictions as mentioned above, the transcription process can also be integrated with the Transformer-based language model through the context beam search algorithm [65]. The Transformer-based language model follows the architecture specified in [65] and is trained with the same corpora as the tri-gram language model using the Fairseq [66] toolkit.…”

Section: Transcriptionmentioning

confidence: 99%

“…It can be seen that our approach achieves state-of-the-art performance with and without language model. Specifically, taking advantage of the Transformer-based language model, the method in [65] outperforms our approach that uses the traditional trigram language model. However, state-of-the-art performance can still be achieved when our method is also equipped with the Transformer-based language model.…”

Section: F Experiments On Icdar2013-offline 1) Experimental Settingsmentioning

confidence: 99%

See 4 more Smart Citations

Recognition of Handwritten Chinese Text by Segmentation: A Segment-Annotation-Free Approach

Peng

Jin

et al. 2023

IEEE Trans. Multimedia

View full text Add to dashboard Cite

Online and offline handwritten Chinese text recognition (HTCR) has been studied for decades. Early methods adopted oversegmentation-based strategies but suffered from low speed, insufficient accuracy, and high cost of character segmentation annotations. Recently, segmentation-free methods based on connectionist temporal classification (CTC) and attention mechanism, have dominated the field of HCTR. However, people actually read text character by character, especially for ideograms such as Chinese. This raises the question: are segmentation-free strategies really the best solution to HCTR? To explore this issue, we propose a new segmentation-based method for recognizing handwritten Chinese text that is implemented using a simple yet efficient fully convolutional network. A novel weakly supervised learning method is proposed to enable the network to be trained using only transcript annotations; thus, the expensive character segmentation annotations required by previous segmentation-based methods can be avoided. Owing to the lack of context modeling in fully convolutional networks, we propose a contextual regularization method to integrate contextual information into the network during the training stage, which can further improve the recognition performance. Extensive experiments conducted on four widely used benchmarks, namely CASIA-HWDB, CASIA-OLHWDB, ICDAR2013, and SCUT-HCCDoc, show that our method significantly surpasses existing methods on both online and offline HCTR, and exhibits a considerably higher inference speed than CTC/attention-based approaches.

show abstract