An Optimal Thresholder for Ancient Degraded Palm leaf Document Binarization

This paper presents a novel adaptive thresholding algorithmic model for ancient document image binarization. The objective of the algorithm is mainly to perform the removal of ink bleed troughs, smears, smudges and non-uniform illumination in palm leaf documents and also on degradation such as reflection of text incurred from other side of two-sided degraded text document images as well as palm leaf, Algorithm transmits the document image through an established sequence of steps such as contrast stretching transforms, conversion to greyscale image, a heuristic adaptive thresholder and post enhancement protocols. Experimentations are conducted on various bench marking ancient data-sets of DIBCO 2009, DIBCO 2010, DIBCO 2011, DIBCO 2012, DIBCO 2013, DIBCO 2014, DIBCO, 2016, DIBCO 2017, DIBCO 2018 and 30 palm leaf thanthrashashtam documents made available as a part of document image binarization contest. The results prove the efficiency of our algorithm towards handling variety of degradations with respect to DIBCO data-sets as well as palm leaf documents with its simplicity carried out through objective evaluation methods. A better PSNR of more than 10 is obtained for more than 80% of both documents and validation of enhanced documents with ground truths produced an average precision and sensitivity of about 95% for all the document samples tested.