Advanced Image Preprocessing Strategies For Mitigating Challenges In Optical Character Recognition Of Diverse Textual Images

Abstract

Optical Character Recognition (OCR) is an innovative technology that transforms text in images into text that machines can read. This process is essential for digitizing documents, streamlining data entry, and facilitating text searches within scanned files. However, OCR accuracy is often hindered by poor image quality, noise, and varying text formats, necessitating advanced preprocessing techniques to optimize text clarity and contrast. This study focuses on enhancing OCR accuracy through improved image preprocessing, employing three key methods: Warp Perspective for accurate alignment and cropping of quadrilateral planes; Gaussian Blurring to reduce noise and enhance the distinction between text and background; and Otsu Binarization to convert images into binary format with optimal thresholding. Results indicate significant improvements, with printed text accuracy increasing from 65.56% to 90.35% and mixed text accuracy rising from 41.15% to 56.31%, while handwritten text showed modest gains from 12.27% to 21.08%, reflecting its inherent complexity. These findings demonstrate the potential of advanced preprocessing to enhance OCR performance, particularly for printed and mixed text, while highlighting the need for specialized techniques or models for handwriting recognition. Future research could explore automated cropping using contour detection to improve efficiency and adaptability to diverse document sizes.