OCR post-correction for detecting adversarial text images

Imam, Niddal H.; Kolovos, Dimitrios S.

doi:10.1016/j.jisa.2022.103170

Cited by 7 publications

(2 citation statements)

References 44 publications

(107 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The process of digitizing text and characters has become easier with the advent of Optical Character Recognition (OCR) platforms, which can electronically identify and convert text or print documents into digital text documents [50]- [52]. OCR has become the first layer in text or characters [53] and is commonly used by companies and industries to automate the processing of managing physical typewritten documents, such as creating digital copies of structured documents like invoices, receipts, bank statements, and other accounting documents that need to be managed.…”

Section: Industrial Practice On Ed Analysismentioning

confidence: 99%

A Review of Neural Network Approach on Engineering Drawing Recognition and Future Directions

Mohd Yazed,

Ahmad Shaubari,

Yap

2023

JOIV : Int. J. Inform. Visualization

View full text Add to dashboard Cite

Engineering Drawing (ED) digitization is a crucial aspect of modern industrial processes, enabling efficient data management and facilitating automation. However, the accurate detection and recognition of ED elements pose significant challenges. This paper presents a comprehensive review of existing research on ED element detection and recognition, focusing on the role of neural networks in improving the analysis process. The study evaluates the performance of the YOLOv7 model in detecting ED elements through rigorous experimentation. The results indicate promising precision and recall rates of up to 87.6% and 74.4%, respectively, with a mean average precision (mAP) of 61.1% at IoU threshold 0.5. Despite these advancements, achieving 100% accuracy remains elusive due to factors such as symbol and text overlapping, limited dataset sizes, and variations in ED formats. Overcoming these challenges is vital to ensuring the reliability and practical applicability of ED digitization solutions. By comparing the YOLOv7 results with previous research, the study underscores the efficacy of neural network-based approaches in handling ED element detection tasks. However, further investigation is necessary to address the challenges above effectively. Future research directions include exploring ensemble methods to improve detection accuracy, fine-tuning model parameters to enhance performance, and incorporating domain adaptation techniques to adapt models to specific ED formats and domains. To enhance the real-world viability of ED digitization solutions, this work highlights the importance of conducting testing on diverse datasets representing different industries and applications. Additionally, fostering collaborations between academia and industry will enable the development of tailored solutions that meet specific industrial needs. Overall, this research contributes to understanding the challenges in ED digitization and paves the way for future advancements in this critical field.

show abstract

Section: Industrial Practice On Ed Analysismentioning

confidence: 99%

A Review of Neural Network Approach on Engineering Drawing Recognition and Future Directions

Mohd Yazed,

Ahmad Shaubari,

Yap

2023

JOIV : Int. J. Inform. Visualization

View full text Add to dashboard Cite

show abstract

“…Finally, identify the title and paragraph of the image after removing the table and the illustration. Although the serial classification and recognition method has a certain degree of decline in speed, this method solves the problem of competition 1 Introduction Document layout analysis plays an important role in image segmentation, image classification [1], knowledge extraction [2,3], OCR text extraction [4], and digitization of paper documents [5]. The purpose of document layout analysis is to identify and segment relevant attributes such as tables, illustrations, headings and paragraphs in document images.…”

mentioning

confidence: 99%

Serial Document Layout Analysis Method Based on Improved Mask R-CNN Architecture

Zhao

Zhang

2023

Preprint

View full text Add to dashboard Cite

Document Layout Analysis plays an important part in understanding the content of the document and extracting OCR text. Existing methods rely primarily on parallel analytical and identification techniques. Based on the analysis of the advantages and disadvantages of parallel methods, this article proposes a Serial Classification Recognition Method based on the improved architecture of the Mask R-CNN. To be specific, an improved feature pyramid model SCBAM-FPN is proposed, which utilizes Spatial Attention Module and Channel Attention Module as well as Atrous Convolution to reduce the loss of local information in feature maps and expand the receptive field, thereby increasing the performance of FPN express ability. Experiments show that using SCBAM-FPN to replace the original feature pyramid network in Mask R-CNN improves the detection performance of Mask R-CNN network for multi-scale targets. In addition, in order to improve the accuracy of segmentation, a Mask edge correction algorithm is proposed, which combines the FCN semantic segmentation results with edge superpixel information and uses the Mask-IoU mechanism to perform edge corrections on Masks with scores of less than 90%. The serial classification recognition method utilizes the improved Mask R-CNN architecture to perform Document Layout Analysis. Firstly, the document image is recognized and the table is segmented. Secondly, the image after the split table is used to identify and segment the illustrations. Finally, identify the title and paragraph of the image after removing the table and the illustration. Although the serial classification and recognition method has a certain degree of decline in speed, this method solves the problem of competition among multiple attributes and improves the accuracy of recognition compared with existing methods. In this article, experiments that were conducted on three datasets show that the proposed method surpasses existing methods in terms of accuracy and efficiency.

show abstract

A real-time price recognition system using lightweight deep neural networks on mobile devices

Peker,

Turan,

Özkan

et al. 2024

TinyML for Edge Intelligence in IoT and LPWAN Networks

View full text Add to dashboard Cite

OCR post-correction for detecting adversarial text images

Cited by 7 publications

References 44 publications

A Review of Neural Network Approach on Engineering Drawing Recognition and Future Directions

A Review of Neural Network Approach on Engineering Drawing Recognition and Future Directions

Serial Document Layout Analysis Method Based on Improved Mask R-CNN Architecture

A real-time price recognition system using lightweight deep neural networks on mobile devices

Contact Info

Product

Resources

About