A Deep Learning Approach for Digitization of Invoices

Digitizing documents using Optical Character Recognition (OCR) has significantly improved in recent years. It is driven by the increasing demand for digitizing printed documents, especially invoices. The task of extracting data from invoices is now easier using an OCR engine like Tesseract. Giving the invoice as input directly to an OCR doesn’t help maintain its structures, especially if the invoice has tables. Maintaining the invoice structure is also necessary if the data in the table needs to be processed in a structured form like an Excel sheet.This paper proposes a solution to this problem of invoice processing using the invoice segmentation method involving Convolutional Neural Networks (CNNs) and processing the segmented invoice separately with image pre-processing techniques and OCR. It also discusses the importance of deep learning in invoice processing, Tesseract, the need for image pre-processing, and some techniques that can help enhance the accuracy of invoice processing.