Publicação
Image Segmentation and Classification using Small Data: An Application on Friendly Statements - A Computer Vision and Document Intelligence Project
| Resumo: | Insurance companies face significant challenges in managing numerous physical documents containing critical information, resulting in considerable time and cost expenditures. Although Deep Learning models offer a promising solution, their implementation costs and data privacy concerns restrict widespread adoption, especially when dealing with confidential documents. This internship report presents a novel approach to address these challenges by developing a lightweight computer vision solution for accurately detecting and processing checkboxes from Portuguese friendly statements. The key objective was to demonstrate the feasibility of achieving high accuracy without relying on advanced Deep Learning techniques. By leveraging a small set of examples, we successfully extracted checkbox information while mitigating the high computational requirements associated with traditional Deep Learning models. The results highlight the practicality and cost-effectiveness of our approach, offering insurance companies a viable solution to streamline document management, enhance data security, and improve overall efficiency. This research contributes to the computer vision field by providing valuable insights into alternative methodologies that can be adopted to overcome the limitations of Deep Learning, facilitating broader accessibility and utilization among insurance providers. |
|---|---|
| Autores principais: | Gomes, Gonçalo Nuno Matos |
| Assunto: | Computer vision Deep Learning Image segmentation Classification Small data SDG 8 - Decent work and economic growth SDG 9 - Industry, innovation and infrastructure SDG 11 - Sustainable cities and communities SDG 17 - Partnerships for the goals |
| Ano: | 2023 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | Insurance companies face significant challenges in managing numerous physical documents containing critical information, resulting in considerable time and cost expenditures. Although Deep Learning models offer a promising solution, their implementation costs and data privacy concerns restrict widespread adoption, especially when dealing with confidential documents. This internship report presents a novel approach to address these challenges by developing a lightweight computer vision solution for accurately detecting and processing checkboxes from Portuguese friendly statements. The key objective was to demonstrate the feasibility of achieving high accuracy without relying on advanced Deep Learning techniques. By leveraging a small set of examples, we successfully extracted checkbox information while mitigating the high computational requirements associated with traditional Deep Learning models. The results highlight the practicality and cost-effectiveness of our approach, offering insurance companies a viable solution to streamline document management, enhance data security, and improve overall efficiency. This research contributes to the computer vision field by providing valuable insights into alternative methodologies that can be adopted to overcome the limitations of Deep Learning, facilitating broader accessibility and utilization among insurance providers. |
|---|