A method of adaptive binarization of halftone images with additional processing on use of signal-to-noise ratio

• Дата: 2020
• Автори: Egorov, P. M., Yakovchenko, O. I.
• Формат: Стаття
• Мова: Ukrainian
• Опубліковано: Інститут проблем реєстрації інформації НАН України 2020
• Теми: бінаризація зображень документів; вейвлет-пере-творення; кореляційне зіставлення зображень; відношення сигнал/шум; document image binarization; wavelet transform; digital image correlation; signal-to-noise ratio
• Онлайн доступ: http://drsp.ipri.kiev.ua/article/view/211268
• Назва журналу: Data Recording, Storage & Processing

A methodology for binarization of grayscale image of digitized documents is proposed. The process of binarization consists of two main parts. Firstly, one should elicit basic components of grayscale image and calculate their numeric values. Secondly, it is necessary to define the quality of a grayscale image, depending on classification by calculated basic components’ values. The result of elicitation of basic components of grayscale image is the image binarization and it is processed iteratively by three steps of approximation. The first step concludes in using wavelet analytics. Namely, one should define the most valuable parts of image and then, apply image wavelet filtration depending on the most valuable levels of decomposition for image reconstruction.The binarization image of the first step approximation is the achievement grounded on the rule that, if the value of the reconstructed pixel is less than zero, than it belongs to the valuable part of grayscale image. The definition of most valuable levels of wavelet transformation depends on the mathematical apparatus of wavelet packet transformation, which is usually used for image compression. Log-energy wavelet entropy is used as a measurement of level to be accepted as valuable. Uniqueness of this method is that most valuable levels of decomposition are set as only two biggest scales.The second step of approximation is about deletion of background areas mistakenly taken as valuable part at the first approximation step. The upper limit of lightness for valuable part is a criterion to decide about, and is set to such value that correlation between an achieved binarized image and input grayscale image, reaching its maximum. The third step of approximation is the reconstruction of graphic objects which sizes are much bigger than size of wavelets for most valuable levels of decomposition. The final image processing is done with signal-to-noise ratio classification.In this perspective the necessary software was constructed and the results gained with it show that this method makes possible to binarize a wide range of images of documents, including images with low-level of contrast or with signs of fading. Fig.: 8. Refs: 18 titles.