Image recognition expense extraction

Kok, Wei Jin (2021) Image recognition expense extraction. Final Year Project, UTAR.

Preview

PDF Download (2884Kb) | Preview

Abstract

This is a mobile application development project developed for academic purposes. The topics covered are mobile development and OCR. Keeping track of income and expenses both in the short and long term is integral for long-term financial growth, as evidenced by the resources allocated for income and expense tracking in large organizations. Both accounting staff as well as personal assistants to managers may perform resource tracking work. In order to achieve long-term financial goals, families may also want to keep track of financial resources. However, while it may be an essential behaviour for long-term financial growth, income and expense tracking is generally a behaviour that takes effort and discipline. All parties can benefit if the effort and discipline required for tracking is lessened through the development of this application. The examined research includes discussion on the suitability of OCR and Spectral clustering, as well as the pre-processing steps before using Spectral clustering, alongside proposed improvements. Research on training neural networks using transformation and training data of deformed receipt images has examined. The report also details the process of how deformations may be synthetically added to the training to the training dataset, which type of neural network is trained to remove deformations from receipts, and how the receipts may be further processed. The proposed methodology is rapid application development (RAD). The four deliverables of each phase include: a list of functional and non-requirements, a sequence diagram, application iterations, as well as the complete application.

Actions (login required)