Restoring Degraded Receipt Images with Generative Approach

Abstract

Optical Character Recognition (OCR) technologies play a crucial role in processes such as document processing and expense management. Since receipts can easily fade, OCR-based automation often leads to the need for manual corrections. This results in additional time spent by employees and extra costs. Without proper oversight, these issues can cause significant problems in accounting. Therefore, we propose a model to correct degraded documents using a generative AI framework based on transformer architecture. The experiments show that the proposed model provides significant improvements and enhances OCR results.