Enhancement of Degraded Historical Document Images for Binarization

Abstract

This work introduces a technique for enhancing deteriorated images of historical documents, overcoming noise, uneven illumination, and texture degradation through five steps: preprocessing, texture enhancement, illumination correction, binarization, and post-processing. In the preprocessing phase, Wiener filtering is employed to effectively reduce noise and enhance overall image quality. Subsequently CLAHE is used to improve local contrast and detail, while Local Binary Patterns (LBP) further enhance texture patterns, contributing to the overall enhancement of image quality. To mitigate the effects of uneven illumination, Retinex-based algorithms are utilized following texture enhancement, ensuring preservation of texture while correcting illumination variations across the image. The binarization step employs Sauvola's method, segmenting the image into foreground and background regions based on local pixel intensity variations. Post-processing techniques, including morphological methods, are then applied to eliminate minor artifacts and refine text shapes in the binarized image, thereby improving readability. Experimental testing on a series of DIBCO datasets confirms the suggested methodology's capacity to considerably improve the visual quality and readability of historical document images. The suggested method outperforms existing methods in terms of metrics such as F-measure, PSNR, NRM and DRD, emphasizing its potential for advancing research in document restoration and analysis.