Character Recognition Performance Research Articles

A-182 Applications of artificial intelligence optical character recognition in laboratory clerical functions offsetting staffing shortages and error reduction

Abstract Background The clinical laboratory profession has been dealing with a decade long, and now critical retention and staffing problem. The average national turnover rate for laboratory technicians is 17%, and the highest turnover rate exhibited for phlebotomists is 36%. While turnover rates have an impact on a laboratory, or hospitals budget, more importantly they have a substantial impact on quality and error rates, particularly in the pre-analytical stage of testing; with studies showing up to 65.09% of errors impacting the pre-analytical phase. Methods In an effort to remedy this high error rate, a meta-analysis was performed on paper order entry and clerical pre-analytical error, in parallel with performance and accuracy data of advanced optical character recognition driven by artificial intelligence and machine learning (AI OCR). Results AI OCR studies demonstrated an untrained accuracy percentage, to include handwriting of 91.08%, and when trained utilizing both relational and hierarchal logic data bases accuracy increased to 98.4%. The rate at which AI OCR processed and classified medical information was 3-4 pages per second. It is estimated that 99.9%-100% accuracy can be achieved with additional form field training and mapping. The level of accuracy and efficiency for laboratory paper order-entry utilizing AI OCR can also be increased with the addition of natural language processing and data recognition algorithms when combined with logic databases containing information such as medical record numbers, ICD-10 codes, and laboratory test compendiums. Conclusions The performance of this technology demonstrates a strong application to reduce the burden of clerical paper order entry, while addressing the 30.8% dissatisfaction rate regarding manual entry processes among lab professionals surveyed. Moreover, AI OCR clerical data classification, extraction and entry processes augmented with human verification steps examined, demonstrated overall accuracy of 99.99%. The application of AI OCR in laboratory pre-analytical processes has the potential to improve test order accuracy and efficiency, ultimately impacting the quality of patient care and operational costs for healthcare facilities. When examined through the lens of staff retention, studies show that 80% of laboratory employees feel that a workplace with the necessary tools and technology positively impacts their job satisfaction. While organizations investing in technologies that drive both job efficiency and satisfaction see a 125% return on investment through productivity. Demonstrating that the application of sophisticated technologies that both increase productivity and decrease error rates have a substantial impact on job satisfaction and ultimately retention. Although laboratory clerical data and order entry processes were examined for this meta-analysis, the aggregate data compiled showed that the application of AI OCR in clerical processes could also have a substantial impact on the 25.9% annual turnover rate for hospitals as a whole in the United States.

Incomplete handwritten Dongba character image recognition by multiscale feature restoration

Incomplete handwritten Dongba character often appears in heritage documents and its recognition is significant for heritage and philology. However, all previous methods always suppose that a complete Dongba character is used as input, and thus fail to achieve satisfactory performance when applied to incomplete Dongba character recognition. In this paper, an end-to-end network (DB2RNet) is proposed for incomplete handwritten Dongba character image recognition by multiscale feature restoration. Specifically, we first develop datasets that contain different levels of incomplete Dongba characters. A restoration module is proposed to restore the input incomplete Dongba character, and then a recognition module is employed to recognize Dongba character. By introducing an inter-module residual connection between the restoration module and recognition module, the DB2RNet can strengthen feature information transmission and boost the recognition performance. In addition, novel multiscale feature blocks are introduced, which can provide more effective texture and contextual information transmission for Dongba character image restoration, and thus yielding better restoration effects and better recognition results. Extensive experiments are conducted on Dongba character, Chinese character and Oracle character datasets and validate the effectiveness, superiority and robustness of our methods. Experiments results demonstrate that our proposed DB2RNet achieves competitive Dongba character restoration and recognition performance and outperforms the current state-of-the-art methods.

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.

Optimizing OCR Performance for Programming Videos: The Role of Image Super-Resolution and Large Language Models

The rapid evolution of video programming tutorials as a key educational resource has highlighted the need for effective code extraction methods. These tutorials, varying widely in video quality, present a challenge for accurately transcribing the embedded source code, crucial for learning and software development. This study investigates the impact of video quality on the performance of optical character recognition (OCR) engines and the potential of large language models (LLMs) to enhance code extraction accuracy. Our comprehensive empirical analysis utilizes a rich dataset of programming screencasts, involving manual transcription of source code and the application of both traditional OCR engines, like Tesseract and Google Vision, and advanced LLMs, including GPT-4V and Gemini. We investigate the efficacy of image super-resolution (SR) techniques, namely, enhanced deep super-resolution (EDSR) and multi-scale deep super-resolution (MDSR), in improving the quality of low-resolution video frames. The findings reveal significant improvements in OCR accuracy with the use of SR, particularly at lower resolutions such as 360p. LLMs demonstrate superior performance across all video qualities, indicating their robustness and advanced capabilities in diverse scenarios. This research contributes to the field of software engineering by offering a benchmark for code extraction from video tutorials and demonstrating the substantial impact of SR techniques and LLMs in enhancing the readability and reusability of code from these educational resources.

Build a Trained Data of Tesseract OCR engine for Tifinagh Script Recognition

This article introduces a methodology for constructing a trained dataset to facilitate Tifinagh script recognition using the Tesseract OCR engine. The Tifinagh script, widely used in North Africa, poses a challenge due to the lack of built-in recognition capabilities in Tesseract. To overcome this limitation, our approach focuses on image generation, box generation, manual editing, charset extraction, and dataset compilation. By leveraging Python scripting, specialized software tools, and Tesseract's training utilities, we systematically create a comprehensive dataset for Tifinagh script recognition. The dataset enables the training and evaluation of machine learning models, leading to accurate character recognition. Experimental results demonstrate high accuracy, precision, recall, and F1 score, affirming the effectiveness of the dataset and its potential for practical applications. The results highlight the robustness of the OCR system, achieving an outstanding accuracy rate of 99,97 %. The discussion underscores its superior performance in Tifinagh character recognition, exceeding the findings in the field. This methodology contributes significantly to enhancing OCR technology capabilities and encourages further research in Tifinagh script recognition, unlocking the wealth of information contained in Tifinagh documents

Deep learning-based text detection and recognition on architectural floor plans

An important aspect of automatic floor plan analysis is the extraction of textual information, as it is essential for a thorough understanding of the drawing. This paper presents a text extraction approach utilizing a deep learning-based object detection model and state-of-the-art Optical Character Recognition (OCR) methods. The paper contributes to the research community in three ways: First, it introduces additional annotations to existing data sets to encompass text elements. Second, it proposes a specialized data synthesis pipeline, allowing for generating training images that mimic important characteristics of real data. Finally, it documents a comparative study of deep learning-based object detection architectures (Tesseract, EAST, CRAFT, Faster R-CNN, YOLOv5, YOLOR, YOLOv7, and YOLOv8) and OCR tools (PARSEq, MATRN, EasyOCR, and Tesseract) for the task. Results indicate that YOLOv7 yields the best text detection performance (up to 97.5% wmAP) and PARSEq excels in character recognition (85.2% CER). The data sets are made available.

Crossmodal transfer of perceptual learning: Evidence from the recognition of first and second language characters in young children.

Crossmodal transfer of learning is a neurocognitive process whereby a learner's experience and knowledge acquired through one sensory mode enable him/her to perform a similar task using a different sensory mode. This study examined the transfer of (mostly) visually acquired knowledge of first- and second-language characters to the tactile modality typically not used in that acquisition process. Two experiments were conducted, one to assess letter recognition skills and one to assess digit recognition skills in both Bangla and English, in 30 sighted young children who had mastered those characters through sensory learning in natural settings. Results unequivocally demonstrated that children were able to recognize/classify the first and second language letters or digits presented not only to the (trained) visual modality but to the (untrained) tactile modality as well, and as expected, with greater recognition accuracy and shorter recognition time in the former than the latter modality. Their character recognition performance was found to be significantly influenced not by language but by character type, with digits being more accurately and more speedily recognized than letters. Moreover, language-task modality interaction was found to mediate letter recognition accuracy, digit recognition accuracy, and digit recognition time, whereas character type-task modality interaction was found to significantly mediate character recognition time only. The ecological and theoretical significance of these findings is discussed. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

A novel multiclass-based framework for P300 detection in BCI matrix speller: Temporal EEG patterns of non-target trials vary based on their position to previous target stimuli

Brain–computer interface (BCI) provides a new communication pathway for severely disabled people and enables them to communicate with external world using only their brain activity. P300-based BCI speller helps patients spell words using their brain signals.Until now, binary-classification-based approaches have been used for P300 detection. This study demonstrates that binary-classification-based approaches may not be appropriate for P300 detection. We proved that temporal EEG patterns of non-target trials are different based on their position to previous target stimuli. Therefore, considering all non-target trials in only one group makes distinguishing target from non-target components difficult for machine learning algorithms and, consequently, deteriorate character recognition accuracy. This study introduced a novel approach for P300 detection in BCI spellers. In this study, we first divided non-target trials into several groups according to their temporal patterns in training stage. Then, we introduced a multiclass-based framework for P300 detection.Proposed approach is evaluated with three public datasets, BCI competition II, BCI competition III and the BNCI Horizon. In all three datasets, proposed multi-class approach outperformed common binary-classification-based approach in the same conditions. In addition, multiclass-based approach reached the highest accuracy (100%) in the 3th sequence for BCI competition II, for BCI competition III achieved an average accuracy of 74% and 98% in 5th and 15th sequences and, for the BNCI Horizon dataset achieved an average accuracy of 77.50% and 97.49% in 5th and 10th sequences, respectively. That means our proposed approach, regardless of its simplicity, achieved state-of-the-art character recognition performance in the existing methods. The results confirmed that binary-classification based methods is not appropriate for P300 detection in BCI spellers.

Towards better long-tailed oracle character recognition with adversarial data augmentation

Deciphering oracle bone script is of great significance to the study of ancient Chinese culture as well as archaeology. Although recent studies on oracle character recognition have made substantial progress, they still suffer from the long-tailed data situation that results in a noticeable performance drop on the tail classes. To mitigate this issue, we propose a generative adversarial framework to augment oracle characters in the problematic classes. In this framework, the generator produces synthetic data through convex combinations of all the available samples in the corresponding classes, and is further optimized through adversarial learning with the classifier and simultaneously the discriminator. Meanwhile, we introduce Repatch to generalize samples in the generator. Since tail classes do not have sufficient data for convex combinations, we propose the TailMix mechanism to generate suitable tail class samples from other classes. Experimental results show that our proposed algorithm obtains remarkable performance in oracle character recognition and achieves new state-of-the-art average (total) accuracy with 86.03% (89.46%), 86.54% (93.86%), 95.22% (96.17%) on the three datasets Oracle-AYNU, OBC306 and Oracle-20K, respectively.

Multiple Scale Convolutional Few-Shot Learning Networks for Online P300-Based Brain–Computer Interface and Its Application to Patients With Disorder of Consciousness

P300 brain-computer interfaces (BCIs) have significant potential for detecting and assessing residual consciousness in patients with disorders of consciousness (DoC) but are limited by insufficient data collected from them. In this study, a multiple scale convolutional few-shot learning network (MSCNN-FSL) was proposed to detect and recognize small sample P300 signals. A multiple scale convolutional neural network (MSCNN) was developed to learn different scale features from different scales of receptive fields to obtain more information from electroencephalograms (EEG). Then, a prototypical network with cosine distance was introduced as a classifier to classify and small sample P300 signals. The MSCNN-FSL was evaluated in two independent online BCI experiments. In the first P300 speller experiment, the presented network achieves good character recognition performance with average accuracies of 98.02%±1.70%. In the second experiment, eight healthy controls achieved photo recognition performance with average accuracies of 98.75%±1.49% and three of the twelve DoC patients achieved more than 64% online accuracies with significance. Our results indicated that the proposed MSCNN-FSL could correctly assess the three patients who may have residual consciousness but were misdiagnosed by the Coma Recovery Scale-Revised (CRS-R). Clinical evaluation after two months did prove our BCI assessment results for the three patients.

ESRGAN 기반 영상 초해상화를 통한 전기설비 상태정보 추출 연구

Electrical equipment performs external communication for monitoring. When a communication malfunction occurs, there is a need for a way to extract information from the electrical equipment. One approach is to capture the display of electrical equipment with an image using devices like CCTV and then extract the information from the image using optical character recognition. However, the images are low-resolution, so the optical character recognition does not work well on the image. This paper proposes a simple method to improve the performance of optical character recognition with a super-resolution model. The proposed method converts the low-resolution image to a high-resolution image through the super-resolution model trained with a proper electrical equipment image dataset. As a result, optical character recognition can extract information from high-resolution images. Experiments on a real-world electrical equipment image dataset show that the proposed method helps to extract information from electrical equipment images.

Neural Correlates of Handwriting Effects in L2 Learners.

Learning to write involves integrating motor production and visual perception to develop orthographic representations. This study tries to test the effect of hand movement training as a pathway to neural correlates for L2 Chinese and L2 English readers. Twenty L2 Chinese and 20 L2 English (n = 20) adults participated in both behavioral and electroencephalogram (EEG) experiments. We designed six learning conditions: Hand Writing Chinese (HC), Viewing Chinese (VC), Drawing followed by Character Recognition in Chinese (DC), Hand Writing English (HE), Viewing English (VE), and Drawing followed by Word Recognition in English (DE). Behavioral and EEG results demonstrated that drawing facilitated visual word recognition in Chinese compared to viewing. The findings imply that hand movement could strengthen the neural processing and improve behavioral performance in Chinese character recognition for L2 Chinese learners and English word recognition for L2 Chinese learners. Furthermore, N170 amplitude at the drawing condition was positively correlated with N400 amplitudes. Thus, the early visual word recognition neural indicator (e.g., N170) was predictive of the late neural indicator of semantic processing (e.g., N400), suggesting that hand movement facilitates the neural correlates between early word recognition and later comprehension.

IMPROVED OCR QUALITY FOR SMART SCANNED DOCUMENT MANAGEMENT SYSTEM

The quality of the document images is a crucial factor for the performance of an Optical Character Recognition (OCR) model. Various issues from the input data hinder the recognition success such as heterogeneous layouts, skewness and proportional fonts. This paper investigated several algorithms for data pre-processing including image deskewing, table and document layout analysis to improve the accuracy of the OCR model and then built an end-to-end scanned document management system. We verified the algorithms using a well-known OCR software namely Tesseract. The experiments on a real dataset shown that our methods can accurately process document images with arbitrary angles of rotation, and different layouts. As a result, the accuracy by words of Tesseract can boost 23% for documents with complex structures. The quality of the output text allows to build a system to store and search documents efficiently. Index

An Interpretable Convolutional Neural Network for P300 Detection: Analysis of Time Frequency Features for Limited Data

In this study, a new deep learning-based methodology is developed for P300 detection in brain computer interface (BCI) systems based on time-frequency (TF) features of EEG signals coupled to deep learning. The TF distributions can transform EEG signals to the TF images by simultaneous representation of time and frequency properties of the signal. However, they do not display the energy distribution of signals at different scales identically and their advantages may be greatly incorporated by using them together. Here, four TF images of single-trail EEG signal are computed and the concatenation of the TF (cTF) images of each signal is developed to be used as training data for a simple and lightweight deep learning-based classifier. The applied TF distributions are spectrogram, Wigner&#x2013;Ville distribution, Morlet-scalogram, and Bertrand distribution. Performance of method is evaluated over limited data acquired from the normal and amyotrophic lateral sclerosis (ALS) datasets and accuracy of 96.56&#x0025;, and 96.84&#x0025; are achieved respectively, which is superior to the other comparing algorithms. Moreover, results of cross-subject classification indicate the promising ability of the method in eliminating calibration in BCI systems. Furthermore, the heat maps of the P300 and non-P300 classes are produced to explain important regions of cTF image for classifier decision and investigate which TF may help better classification. Results revealed the efficiency of cTF images for accurate P300 detection in simple structure classifiers having the advantage of fewer data and less memory requirement. This method can be employed in P300 speller BCI systems to improve the character recognition performance in.

Gradient-Free Neural Network Training via Synaptic-Level Reinforcement Learning

An ongoing challenge in neural information processing is the following question: how do neurons adjust their connectivity to improve network-level task performance over time (i.e., actualize learning)? It is widely believed that there is a consistent, synaptic-level learning mechanism in specific brain regions, such as the basal ganglia, that actualizes learning. However, the exact nature of this mechanism remains unclear. Here, we investigate the use of universal synaptic-level algorithms in training connectionist models. Specifically, we propose an algorithm based on reinforcement learning (RL) to generate and apply a simple biologically-inspired synaptic-level learning policy for neural networks. In this algorithm, the action space for each synapse in the network consists of a small increase, decrease, or null action on the connection strength. To test our algorithm, we applied it to a multilayer perceptron (MLP) neural network model. This algorithm yields a static synaptic learning policy that enables the simultaneous training of over 20,000 parameters (i.e., synapses) and consistent learning convergence when applied to simulated decision boundary matching and optical character recognition tasks. The trained networks yield character-recognition performance comparable to identically shaped networks trained with gradient descent. The approach has two significant advantages in comparison to traditional gradient-descent-based optimization methods. First, the robustness of our novel method and its lack of reliance on gradient computations opens the door to new techniques for training difficult-to-differentiate artificial neural networks, such as spiking neural networks (SNNs) and recurrent neural networks (RNNs). Second, the method’s simplicity provides a unique opportunity for further development of local information-driven multiagent connectionist models for machine intelligence analogous to cellular automata.

A Comparative Study of Three Measurement Methods of Chinese Character Recognition for L2 Chinese Learners.

Measuring Chinese character recognition ability is essential in research on character learning among learners of Chinese as a second language (CSL). Three methods are typically used to evaluate character recognition competence by investigating the following properties of a given character: (a) pronunciation (phonological method), (b) meaning (semantic method), and (c) pronunciation and meaning (phonological and semantic or PS method). However, no study has explored the similar or dissimilar outcomes that these three measurements might yield. The current study examined this issue by testing 162 CSL learners with various L1 backgrounds and Chinese proficiency levels. Participants' performance in character recognition measured using a phonological method, a semantic method, and a PS method was compared, which led to two major findings. In terms of similarity, participants' performance in character recognition and the influence of L1 background and Chinese proficiency level on character recognition was similar across the three methods. As for differences, the semantic method could yield a character recognition test with better quality than the other two methods, and the three methods yielded different best fitting models and showed different predictions for Chinese proficiency across different L1 groups. Theoretical and practical implications of these findings are proposed.

DeblurGAN-CNN: Effective Image Denoising and Recognition for Noisy Handwritten Characters

Many problems can reduce handwritten character recognition performance, such as image degradation, light conditions, low-resolution images, and even the quality of the capture devices. However, in this research, we have focused on the noise in the character images that could decrease the accuracy of handwritten character recognition. Many types of noise penalties influence the recognition performance, for example, low resolution, Gaussian noise, low contrast, and blur. First, this research proposes a method that learns from the noisy handwritten character images and synthesizes clean character images using the robust deblur generative adversarial network (DeblurGAN). Second, we combine the DeblurGAN architecture with a convolutional neural network (CNN), called DeblurGAN-CNN. Subsequently, two state-of-the-art CNN architectures are combined with DeblurGAN, namely DeblurGAN-DenseNet121 and DeblurGAN-MobileNetV2, to address many noise problems and enhance the recognition performance of the handwritten character images. Finally, the DeblurGAN-CNN could transform the noisy characters to the new clean characters and recognize clean characters simultaneously. We have evaluated and compared the experimental results of the proposed DeblurGAN-CNN architectures with the existing methods on four handwritten character datasets: n-THI-C68, n-MNIST, THI-C68, and THCC-67. For the n-THI-C68 dataset, the DeblurGAN-CNN achieved above 98% and outperformed the other existing methods. For the n-MNIST, the proposed DeblurGAN-CNN achieved an accuracy of 97.59% when the AWGN+Contrast noise method was applied to the handwritten digits. We have evaluated the DeblurGAN-CNN on the THCC-67 dataset. The result showed that the proposed DeblurGAN-CNN achieved an accuracy of 80.68%, which is significantly higher than the existing method, approximately 10%.

Alternative Collaborative Learning for Character Recognition in Low-Resolution Images

Character recognition in a single image is a technology utilized in various sensor platforms, such as smart parking and text-to-speech systems, and numerous studies are being conducted to improve its performance by experimenting with novel approaches. However, when low-quality images were inputted to a character recognition neural network for recognition, a difference in the resolution of the training image and low-quality image results in poor accuracy. To resolve this problem, this study proposes a collaborative trainable mechanism that integrates a global image feature extraction-based super-resolution neural network with a character recognition neural network. This collaborative trainable mechanism helps the character recognizer to be robust to inputs with varying quality in the real world. The alternative collaborative learning and character recognition performance test was conducted using the license plate image dataset among various character images, and the effectiveness of the proposed algorithm was verified using a performance test.

Design of an OCR System and its Hardware Implementation

Abstract: In this paper, we present a scheme to develop to complete OCR system for printed text English Alphabet of Uppercase of different font and of different sizes so that we can use this system in Banking, Corporate, Legal industry and so on. OCR system consists of different modules like preprocessing, segmentation, feature extraction and recognition. In preprocessing step it is expected to include image gray level conversion, binary conversion etc. After finding out the feature of the segmented characters artificial neural network and can be used for Character Recognition purpose. Efforts have been made to improve the performance of character recognition using artificial neural network techniques. The proposed OCR system is capable of accepting printed document images from a file and implemented using MATLAB R2014a version. Key words: OCR, Printed text, Barcode recognition

A new dynamically convergent differential neural network for brain signal recognition

In this paper, a dynamically convergent differential neural network (DCDNN) is proposed for P300 brain signal classification, which can be applied to brain-computer interface (BCI) systems. Being different from most of state-of-the-art neural networks with thousands of parameters to adjust, the proposed neural network possesses fast training rate since it is a lightweight neural network. In the framework, data preprocessing includes time window interception, bandpass filtering, moving average filtering and training set balance. After that, feature extraction based on PCA is used, and an ensemble DCDNN (E-DCDNN) is proposed for classification. The proposed framework can effectively increase signal-to-noise ratio (SNR), reduce classification dimension without distortion, and eliminate classifier variability. In this work, the proposed framework achieves 100% and 98% accuracy on Data Set IIb of BCI Competition II and Data Set II of BCI Competition III, which mean the state-of-the-art character recognition performance in existing methods.