Improving Binarization Methods for Historical Handwritten Documents
David G. Asatryan, Mariam E. Haroutunian, Grigor Sazhumyan, Alexander V. Kupriyanov, IPSI RAS, Парингер Рустам Александрович, IPSI RAS, Dmitriy V. Kirsh, IPSI RAS · 2023
Binarization of historical documents is a rather complex task, which is intensively dealt with by researchers all over the world.A large number of approaches, procedures, and binarization algorithms have been proposed, but methods that work equally well in all cases have not yet been proposed.Various criteria for evaluating the quality of binarization result are proposed in the literature.In the case of the binarization of ancient handwritten texts, the degree of readability of the text by a visual method or using technical means is taken as a criterion for the quality of the binarization algorithm.One of the approaches to improve the quality of the binarization result proposed in the literature is the preliminary processing of the original image, using filtering methods, morphological analysis, spectral analysis, etc., and the selection of segments with certain sizes.The proposed procedure allows selecting objects in the image with certain sizes, in particular, artifacts existing in the binarized image.In this work, the possibility of improving the quality of a binary image obtained by Niblack's method by applying the above-mentioned procedure is investigated experimentally.For comparison, the results of Otsu's and Sauvola's methods are used.