Publication
Artificial Intelligence in Invoice Recognition: a Systematic Literature Review
| dc.contributor.advisor | Ribeiro, António Rui Trigo | |
| dc.contributor.author | Kukharska, Oleksandra | |
| dc.date.accessioned | 2024-01-12T16:22:33Z | |
| dc.date.available | 2024-01-12T16:22:33Z | |
| dc.date.issued | 2023 | |
| dc.description.abstract | In the era marked by a flourishing economy and rapid advancements in information technology, the proliferation of invoice data has accentuated the urgent need for automated invoice recognition. Traditional manual methods, long relied upon for this task, have proven to be inefficient, error-prone, and incapable of coping with the rising volume of invoices. This research endeavours to addresses the imperative of automating invoice recognition by exploring, assessing, and advancing cutting-edge algorithms, techniques, and methods within the domain of Artificial Intelligence (AI). This research conducts a comprehensive Systematic Literature Review (SLR) to investigate Computer Vision (CV) approaches, encompassing image preprocessing, Layout Analysis (LA), Optical Character Recognition (OCR), and Information Extraction (IE). The objective is to provide valuable insights into these fundamental components of invoice recognition, emphasizing their significance in achieving accuracy and efficiency. This exploration aims to contribute to the development of more effective automated systems for extracting information from invoices, addressing the challenges posed by diverse formats and content. The results indicate that in LA, the combination of Mask Region-based Convolutional Neural Networks (M-RCNN) and Feature Pyramid Network (FPN) achieves goods results. In OCR, algorithms like Convolutional Recurrent Neural Network (CRNN), You Only Look Once version 4 (YOLOv4) and models inspired by M-RCNN and Faster Region-based Convolutional Neural Network (F-RCNN) with ResNetXt-101 as the backbone demonstrate remarkable performance. When it comes to IE, algorithms inspired by F-RCNN and Region Proposal Network (RPN), Grid Convolutional Neural Network (G-CNN) and Layer Graph Convolutional Networks (LGCN), and Gated Graph Convolutional Network (GatedGCN) consistently deliver the best results. | pt_PT |
| dc.identifier.tid | 203464273 | pt_PT |
| dc.identifier.uri | http://hdl.handle.net/10400.26/48887 | |
| dc.language.iso | eng | pt_PT |
| dc.subject | Invoice | pt_PT |
| dc.subject | Invoice Recognition | pt_PT |
| dc.subject | Artificial Intelligence | pt_PT |
| dc.subject | Algorithms | pt_PT |
| dc.subject | Computer Vision | pt_PT |
| dc.subject | Data Extraction | pt_PT |
| dc.title | Artificial Intelligence in Invoice Recognition: a Systematic Literature Review | pt_PT |
| dc.type | master thesis | |
| dspace.entity.type | Publication | |
| rcaap.rights | openAccess | pt_PT |
| rcaap.type | masterThesis | pt_PT |
| thesis.degree.grantor | Instituto Politécnico de Coimbra |