Investigation of the process of binarization of images using local values of the threshold

Abstract

The use of global binarization thresholds in image processing does not always give the correct result. This is especially common when processing images with uneven illumination. In some areas of the image, the automatically determined binarization threshold makes it possible to obtain sufficiently well visualized objects, while in other areas, the objects necessary for analysis become "overexposed" or, conversely, "shaded". In cases where it is necessary to localize all objects of interest in the image, binarization plays a very important role, especially in cases where the object of interest contains information that will be used in the next stages of processing. Multi-gradation images can contain many objects of interest, such as car license plates, train car numbers, people's faces, and defects in manufactured products. Each of these cases requires high-quality processing for subsequent recognition. If there are noises on the processed image or the brightness indicators are unevenly distributed, then the binarization process can lead to the loss of important information – the loss of a part of the symbol, the breakage of the object's contour, or, conversely, the emergence of new areas that are mistakenly added to the object of interest – shadows of other objects, dirt on the license plate sign. Therefore, the binarization process requires a very accurate preliminary calibration for all possible shooting conditions – daylight and dark hours of the day, taking into account possible noise (interference in signal transmission), extreme situations (strong hail or rain). In this article, the authors investigate the process of binarization of images with uneven illumination using several local binarization thresholds instead of one global one. It is proposed to check the histograms of the obtained fragments for the number of peaks or "modes". If the histogram of a binarized fragment is single-mode, then the given fragment is not subject to further processing and the binarization threshold on it is defined correctly. The study of the relationship between the binarization threshold and such image parameters as dispersion and smoothness has been carried out. On those fragments where the value of the average brightness measure differs from the average for all fragments, the binarization threshold is determined incorrectly. If you set the threshold value higher, closer to the average value for all fragments, then as a result binarization will be carried out correctly.