Degraded Document Image Binarization using Novel Background Estimation Technique

Over the past few decades, the use of scanned historical document images has increased dramatically, especially with the emergence of online libraries and standard benchmark datasets like DIBCO. The historical documents are usually in very-poor conditions containing noises like large ink stains, bleed-through, liquid spills, uneven-background, spots, faded-ink, weak/thin text that makes the task of binarization very difficult. In this paper, we propose an effective degraded document image binarization algorithm that performs accurate text segmentation. Our method first estimates the background utilizing information from neighboring pixels and filter smoothening. The next step is background subtraction that helps in the compensation of background distortions. The document is segmented using Otsu thresholding, and then we process the image to remove the remaining noise and maximize text content using labelled connected components. Our method outperforms several existing and widely used binarization algorithms on F-measure, PSNR, DRD, and pseudo F-measure when evaluated on H-DIBCO 2016 and H-DIBCO 2018 datasets and can very effectively detect faint characters from a document image.