Enhancing Text Recognition in OCR Systems Through Image Processing with BSRGAN
Fernando Baptistella de Lima, Eraylson Galdino da Silva · 2025
Context: Image enhancement is essential for advancing Optical Character Recognition (OCR), a technology widely applied across various Information Systems (IS) to enable accurate text extraction from scanned documents, IDs, invoices, and other document types. Problem: Despite OCR’s importance, challenges such as noise, variable illumination, and low-resolution scans often compromise recognition quality, leading to distorted and inaccurate results. These issues can impact the reliability and effectiveness of IS. Solution: This study presents a methodology to improve the quality of low-resolution images by combining image filtering techniques with OpenCV, super-resolution using the BSRGAN model, and EasyOCR for character extraction. IS Theory: The research is anchored in the Information Quality theory in IS, addressing the importance of improving input data to enhance system outputs and reliability. Method: The proposed methodology consists of two main stages. First, low-resolution images are processed using the BSRGAN super-resolution model, which enhances image quality for improved OCR performance. Then, the enhanced images are processed by an OCR system to extract and convert characters into text. Validation was conducted on three datasets: Brazilian Identity Document (BID), IIIT 5K-Word, and SVHN, simulating real-world application conditions. Summary of Results: The results demonstrate the proposed methodology’s effectiveness in enhancing OCR accuracy, significantly reducing error rates in various contexts. Contributions and Impact on IS: This work contributes to the IS field by providing a solution that enhances OCR input quality, benefiting academia through advanced image processing research and the industry by enabling more reliable text recognition in practical applications.