Chinese Character Recognition Research Articles

Fine Segmentation of Chinese Character Strokes Based on Coordinate Awareness and Enhanced BiFPN

Considering the complex structure of Chinese characters, particularly the connections and intersections between strokes, there are challenges in low accuracy of Chinese character stroke extraction and recognition, as well as unclear segmentation. This study builds upon the YOLOv8n-seg model to propose the YOLOv8n-seg-CAA-BiFPN Chinese character stroke fine segmentation model. The proposed Coordinate-Aware Attention mechanism (CAA) divides the backbone network input feature map into four parts, applying different weights for horizontal, vertical, and channel attention to compute and fuse key information, thus capturing the contextual regularity of closely arranged stroke positions. The network’s neck integrates an enhanced weighted bi-directional feature pyramid network (BiFPN), enhancing the fusion effect for features of strokes of various sizes. The Shape-IoU loss function is adopted in place of the traditional CIoU loss function, focusing on the shape and scale of stroke bounding boxes to optimize the bounding box regression process. Finally, the Grad-CAM++ technique is used to generate heatmaps of segmentation predictions, facilitating the visualization of effective features and a deeper understanding of the model’s focus areas. Trained and tested on the public Chinese character stroke datasets CCSE-Kai and CCSE-HW, the model achieves an average accuracy of 84.71%, an average recall rate of 83.65%, and a mean average precision of 80.11%. Compared to the original YOLOv8n-seg and existing mainstream segmentation models like SegFormer, BiSeNetV2, and Mask R-CNN, the average accuracy improved by 3.50%, 4.35%, 10.56%, and 22.05%, respectively; the average recall rates improved by 4.42%, 9.32%, 15.64%, and 24.92%, respectively; and the mean average precision improved by 3.11%, 4.15%, 8.02%, and 19.33%, respectively. The results demonstrate that the YOLOv8n-seg-CAA-BiFPN network can accurately achieve Chinese character stroke segmentation.

Neural mechanisms of Chinese character recognition, updating, and maintenance in the N-back task

Using the N-back task, we investigated how memory load influences the neural activity of the Chinese character cognitive subprocess (recognition, updating, and maintenance) in Mainland Chinese speakers. Twenty-seven participants completed the Chinese character N-back paradigm while having their event-related potentials recorded. The study employed time and frequency domain analyses of EEG data. Results showed that accuracy decreased and response times increased with larger N values. For ERPs, N2pc and P300 amplitudes decreased and SW amplitude increased with larger N values. For time frequency analyses, the desynchronization of alpha oscillations decreased after stimulus onset, but the synchronization of alpha oscillations increased during the maintenance phase. The results suggest that greater memory load is related to a decrease in cognitive resources during updating and an increase in cognitive resources during information maintenance. The results of a behavioral-ERP data structural equation model analysis showed that the ERP indicators in the maintenance phase predicted behavioral performance.

The underlying neuropsychological and neural correlates of the impaired Chinese reading skills in children with attention deficit hyperactivity disorder.

Impaired basic academic skills (e.g., word recognition) are common in children with Attention Deficit Hyperactivity Disorder (ADHD). The underlying neuropsychological and neural correlates of impaired Chinese reading skills in children with ADHD have not been substantially explored. Three hundred and two children with ADHD (all medication-naïve) and 105 healthy controls underwent the Chinese language skill assessment, and 175 also underwent fMRI scans (84 ADHD and 91 controls). Between-group and mediation analyses were applied to explore the interrelationships of the diagnosis of ADHD, cognitive dysfunction, and impaired reading skills. Five ADHD-related brain functional networks, including the default mode network (DMN) and the dorsal attention network (DAN), were built using predefined regions of interest. Voxel-based group-wise comparisons were performed. The ADHD group performed worse than the control group in word-level reading ability tests, with lower scores in Chinese character recognition (CR) and word chains (WS) (all P < 0.05). With full-scale IQ and sustained attention in the mediation model, the direct effect of ADHD status on the CR score became insignificant (P = 0.066). The underlying neural correlates for the orthographic knowledge (OT) and CR differed between the ADHD and the control group. The ADHD group tended to recruit more DMN regions to maintain their reading performance, while the control group seemed to utilize more DAN regions. Children with ADHD generally presented impaired word-level reading skills, which might be caused by impaired sustained attention and lower IQ. According to the brain functional results, we infer that ADHD children might utilize a different strategy to maintain their orthographic knowledge and character recognition performance.

Chinese Character Recognition Using Non-negative Matrix Factorization

Non-negative matrix factorization (NMF) was introduced by Paatero and Tapper in 1994 and it was a general way of reducing the dimension of the matrix with non-negative entries. Non-negative matrix factorization is very useful in many data analysis applications such as character recognition, text mining, and others. This paper aims to study the application in Chinese character recognition using non-negative matrix factorization. Python was used to carry out the LU factorization and non-negative matrix factorization of a Chinese character in Boolean Matrix. Preliminary analysis confirmed that the data size of and and are chosen for the NMF of the Boolean matrix. In this project, one hundred printed Chinese characters were selected, and all the Chinese characters can be categorized into ten categories according to the number of strokes , for . The Euclidean distance between the Boolean matrix of a Chinese character and the matrix after both LU factorization and NMF is calculated for further analysis. Paired t-test confirmed that the factorization of Chinese characters in the Boolean matrix using NMF is better than the LU factorization. Finally, ten handwritten Chinese characters were selected to test whether the program is able to identify the handwritten and the printed Chinese characters. Experimental results showed that 70% of the characters can be recognized via the least Euclidean distance obtained. NMF is suitable to be applied in Chinese character recognition since it can reduce the dimension of the image and the error between the original Boolean matrix and after NMF is less than 5%.

Ancient Chinese Character Recognition with Improved Swin-Transformer and Flexible Data Enhancement Strategies.

The decipherment of ancient Chinese scripts, such as oracle bone and bronze inscriptions, holds immense significance for understanding ancient Chinese history, culture, and civilization. Despite substantial progress in recognizing oracle bone script, research on the overall recognition of ancient Chinese characters remains somewhat lacking. To tackle this issue, we pioneered the construction of a large-scale image dataset comprising 9233 distinct ancient Chinese characters sourced from images obtained through archaeological excavations. We propose the first model for recognizing the common ancient Chinese characters. This model consists of four stages with Linear Embedding and Swin-Transformer blocks, each supplemented by a CoT Block to enhance local feature extraction. We also advocate for an enhancement strategy, which involves two steps: firstly, conducting adaptive data enhancement on the original data, and secondly, randomly resampling the data. The experimental results, with a top-one accuracy of 87.25% and a top-five accuracy of 95.81%, demonstrate that our proposed method achieves remarkable performance. Furthermore, through the visualizing of model attention, it can be observed that the proposed model, trained on a large number of images, is able to capture the morphological characteristics of ancient Chinese characters to a certain extent.

Deep Convolutional Neural Networks Based on Knowledge Distillation for Offline Handwritten Chinese Character Recognition

Deep convolutional neural networks (DNNs) have achieved outstanding performance in this field. Meanwhile, handwritten Chinese character recognition (HCCR) is a challenging area of research in the field of computer vision. DNNs require a large number of parameters and high memory consumption. To address these issues, this paper proposes an approach based on an attention mechanism and knowledge distillation. The attention mechanism improves the feature extraction and the knowledge distillation reduces the number of parameters. The experimental results show that ResNet18 achieves a recognition accuracy of 97.63% on the HCCR dataset with 11.25 million parameters. Compared with other methods, this study improves the performance for HCCR.

An approach for handwritten Chinese text recognition unifying character segmentation and recognition

Text line recognition methods can be categorized into explicit segmentation based and implicit segmentation based ones. Explicit segmentation based methods require character-level annotation during training, while implicit segmentation based methods, trained on line-level annotated data, face alignment drift challenges. Though some methods have been proposed to address these challenges using weakly supervised object detection, they often rely on cumbersome pseudo-box generation processes and complex decoding. In this paper, we propose a unified framework to overcome these challenges, achieving high accuracy in text recognition and character segmentation. To eliminate the need of character-level annotated real text line data in training, we introduce a novel training paradigm that utilizes character-level annotated synthetic data and line-level annotated real data jointly. For synthetic data, candidate characters are explicitly aligned with labeled characters to generate hard labels for supervising model training. For real data, implicit alignment is produced by Connectionist Temporal Classification (CTC) mapping to provide soft labels for weakly-supervised model training. And for inference, we propose two decoding strategies leveraging the advantages of Non-Maximum Suppression (NMS) and CTC decoding. Extensive experiments on benchmark datasets demonstrate the superior performance of our method in text recognition and character localization, even with minimal amounts of character-level annotated line data.

Effects of working memory and visual search skill on Chinese reading comprehension: examining the simple view of reading

AbstractTo investigate the longitudinal effects of two domain-general cognitive abilities, namely verbal working memory and visual search skill, on Chinese reading comprehension. To evaluate whether decoding and linguistic comprehension mediate such effects. A total of 202 first-grade Chinese-speaking children from mainland China (Mage = 86 months, SD = 4.9 months; 126 boys) completed tasks that measured verbal working memory, visual search, expressive vocabulary, morphological awareness, Chinese character recognition, and word reading fluency. One year later, they completed a passage comprehension task and a nonverbal intelligence test. Structural equation modeling was conducted with decoding and linguistic comprehension being included as latent variables and mediating factors explaining the effects of working memory and visual search. The children with better verbal working memory and visual search skill exhibited better reading comprehension performance. The effect of working memory on reading comprehension was fully mediated by linguistic comprehension (captured by vocabulary and morphological awareness) but not decoding (measured by word reading accuracy and fluency). However, decoding fully mediated the impact of visual search skill. Our findings suggest that the reading comprehension ability of beginning Chinese readers is associated with verbal working memory and visual search ability. These general cognitive abilities influence reading comprehension through distinct pathways. In line with the simple view of reading theory, decoding and linguistic comprehension skills explain such impacts. This research underlines the value of simultaneously examining general cognitive abilities and exploring their interconnectedness with linguistic skills in the context of reading comprehension.

Large-scale continual learning for ancient Chinese character recognition

Ancient Chinese character recognition is a challenging problem in the field of pattern recognition. It is difficult to collect all character classes during the training stage due to the numerous classes of ancient Chinese characters and the likelihood of discovering new characters over time. A solution to address this problem is continual learning. However, most continual learning methods are not well-suited for large-scale applications, making them insufficient for solving the problem of ancient Chinese character recognition. Although saving raw data for old classes is a good approach for continual learning to address large-scale problems, it is often infeasible due to the lack of data accessibility in reality. To solve these problems, we propose a large-scale continual learning framework based on the convolutional prototype network (CPN), which does not save raw data for old classes. In this paper, several basic strategies have been proposed for the initial training stage to enhance the feature extraction ability and robustness of the network, which can improve the performance of the model in continual learning. In addition, we propose two practical methods in varying feature space (parameters of feature extractor are changeable) and fixed feature space (parameters of feature extractor are fixed), which enable the model to carry out large-scale continual learning. The proposed method does not save the raw data of old classes and enables simultaneous classification of all existing classes without knowing the incremental batch number. Experiments on the CASIA-AHCDB dataset with 5000 character classes demonstrate the effectiveness and superiority of the proposed method.

Developing a web application for Chinese calligraphy learners using convolutional neural network and scale invariant feature transform

Chinese Calligraphy is an ancient art form with practical benefits such as improving elders' cognition and helping diagnose Parkinson's disease. Learning Chinese calligraphy requires a lot of time and effort. This is due to the different styles and the vast number of Chinese characters used daily. Chinese character and style recognition are multi-class classification problems. In multi-class classification problems, the accuracy decreases as the number of classes increases. Previous research on Chinese calligraphy was limited to either style recognition or character recognition. More research that combines style and character recognition is needed. Furthermore, the models in previous research were limited to recognizing small sets of 300 characters. These small sets cover only 12% of the Chinese characters used daily. The small size is a severe limitation for creating applications for Chinese calligraphy learning. Thus, new research is needed on AI models that combine style and character recognition to recognize a much larger dataset, while maintaining reasonable accuracy rates. Besides, there are no applications that use such AI models to help learners of Chinese calligraphy compare their works with those created by skillful calligraphers. This research proposed enhancing Chinese calligraphy recognitions by combining the VGG (Visual Geometry Group)-net architecture, data augmentation techniques, and dropout in the ANN (artificial neural network) hidden layers. The research produced two AI models that can recognize 962 Chinese characters in five calligraphic styles. We proved that using dropout increased the accuracy convergence during training epochs. In the proposed models, style recognition achieved up to 88.3% accuracy, while character recognition achieved 95.6% accuracy. Finally, we built an online learning application by combining the proposed models with Scale Invariant Feature Transform (SIFT). The application could compare the learner's work with skillful calligraphers' and provide a similarity score to help Chinese calligraphy learners track their progress. Five university students tested the application. They provided positive feedback and some suggestions for improvement.

A Review of the Current Status of AI Research in Handwritten Chinese Character Recognition

With the continuous development of artificial intelligence technology, there have been attempts to utilize AI in the field of linguistics. In this context, research and development in AI-based Chinese character recognition technology has spanned 40 years, with its outcomes attracting significant attention. However, in the field of Chinese character recognition, the research into offline handwritten recognition technology is particularly challenging due to the unique characteristics of handwritten Chinese characters. The main issue in current research is the significant decrease in accuracy for characters composed of similar components or characters with similar overall shapes. The focus of research solutions has been on modern mechanical programming and other engineering aspects, while research that integrates the inherent characteristics of Chinese characters themselves has been relatively overlooked. The author believes that combining engineering technology research with a deeper understanding of the nature of Chinese characters can solve many of the current problems. This paper reviews the current state of related domestic and international research to identify areas with significant potential for improvement and progress in research, and finally proposes directions for future research.

The impact of Home Literacy Environment on Chinese children's character recognition, vocabulary knowledge, and reading comprehension: A developmental cascade model

The impact of Home Literacy Environment on Chinese children's character recognition, vocabulary knowledge, and reading comprehension: A developmental cascade model

Research on Offline Handwritten Chinese Character Recognition Algorithm Based on Gated Convolution and Stacked Self-Attention

Research on Offline Handwritten Chinese Character Recognition Algorithm Based on Gated Convolution and Stacked Self-Attention

Online Handwritten Chinese Character Recognition Based on 1-D Convolution and Two-Streams Transformers

Online Handwritten Chinese Character Recognition Based on 1-D Convolution and Two-Streams Transformers

Chinese character recognition with radical-structured stroke trees

Chinese character recognition with radical-structured stroke trees

The Chinese lexicon of deaf readers: A database of character decisions and a comparison between deaf and hearing readers.

We present a psycholinguistic study investigating lexical effects on simplified Chinese character recognition by deaf readers. Prior research suggests that deaf readers exhibit efficient orthographic processing and decreased reliance on speech-based phonology in word recognition compared to hearing readers. In this large-scale character decision study (25 participants, each evaluating 2500 real characters and 2500 pseudo-characters), we analyzed various factors influencing character recognition accuracy and speed in deaf readers. Deaf participants demonstrated greater accuracy and faster recognition when characters were more frequent, were acquired earlier, had more strokes, displayed higher orthographic complexity, were more imageable in reference, or were less concrete in reference. Comparison with a previous study of hearing readers revealed that the facilitative effect of frequency on character decision accuracy was stronger for deaf readers than hearing readers. The effect of orthographic-phonological regularity differed significantly for the two groups, indicating that deaf readers rely more on orthographic structure and less on phonological information during character recognition. Notably, increased stroke counts (i.e., higher orthographic complexity) hindered hearing readers but facilitated recognition processes in deaf readers, suggesting that deaf readers excel at recognizing characters based on orthographic structure. The database generated from this large-scale character decision study offers a valuable resource for further research and practical applications in deaf education and literacy.

SideNet: Learning representations from interactive side information for zero-shot Chinese character recognition

Existing methods for zero-shot Chinese character recognition usually exploit a single type of side information such as radicals, glyphs, or strokes to establish a mapping with the input characters for the recognition of unseen categories. However, these approaches have two limitations. Firstly, the mappings are inefficient owing to their complexity. Some existing methods design radical-level mappings using a non-differentiable dictionary-matching strategy, whereas others construct sophisticated embeddings to map seen and unseen characters into a unified latent space. Although the latter approach is straightforward, it lacks a learnable scheme for explicit structure construction. Secondly, the complementarity within multiple types of side information has not been effectively explored. For example, the radicals provide structural knowledge at an abstract level, whereas glyphs offer detailed information on their figurative counterparts. To this end, we propose a new method called SideNet that jointly learns character-level representations assisted by two types of interactive side information: radicals and glyphs. SideNet contains a structural conversion module that extracts radical knowledge via dimensional decomposition, and a spatial conversion module that encodes the radical counting map to produce an interactive outcome between radicals and glyph. Finally, we propose a new classifier that integrates the converted features by a similarity-guided fusion mechanism. To the best of our knowledge, this study represents the first attempt to integrate these two types of side information and explore a joint representation for zero-shot learning. Experiments show that SideNet consistently outperforms existing methods by a significant margin in diverse scenarios, including handwriting, printed art, natural scenes, and ancient Chinese characters, which demonstrates the potential of joint learning with multiple types of side information.

Analysis of calligraphy Chinese character recognition technology based on deep learning and computer-aided technology

Analysis of calligraphy Chinese character recognition technology based on deep learning and computer-aided technology

Network architecture exploration for Chinese character recognition with densenet

Chinese character recognition can be widely used in many fields. Though Resnet and Densenet are both used in this area already, using these two networks and making a comparison on training performances between them is a ground that has not been explored. In this paper, these two methods are built and compared. Firstly, a dataset of Chinese character including 5,772 images with 28*28 size will be introduced. Next, Resnet and Densenet model pre-trained on the dataset is selected. Then fine-tuning is done to improve the accuracy of networks. After 50 epochs of training, the final result shows that Densenet is more stable compared with Resnet but less efficient with more epoches to perform well.

License plate Chinese character recognition based on ViT model

Transformer applications have been widely used in the computer vision field. Many related literatures show that the advantages of the model such as increased receptive field and globality are gradually emerging in image processing. However, with the popularity of the transformer, whether it can compete with the convolutional neural network (CNN) in terms of performance is still questionable and remains to be further studied. This paper will use the most basic structural model in the visual transformer (ViT) to classify and identify Chinese characters that are frequently used in the field of transportation and logistics and compare them with two classical CNN models. The results demonstrate that the performance of the transformer is obviously better than that of the traditional CNN structure, and the final accuracy of character recognition is higher than that of CNN, up to 98.66 %. It fully shows the infinite potential and excellent performance of the transformer in the area of computer vision and has high reliability and generalization ability.