A systematic approach to the process of character recognition in noisy images

. In the context of recognition of digitized images with low quality due to the presence of various noises, a model of the character recognition process is proposed that allows taking into account the different quality of the input images. The problem of recognizing images containing different types of noise is solved. Algorithms that smooth the image and remove dense accumulation of noise often do not cope with noise scattered evenly throughout the image (“salt”), and the use of several noise suppression algorithms at once can affect the image quality. If you choose the wrong blurring algorithm, you can subsequently get less accurate coordinates of the image contours for further segmentation and recognition when searching for boundaries. When segmenting an image, it may happen that due to incorrect processing, the process of searching for character pixels will lead to an incorrect result. To solve this problem, it is proposed to form a set of blurring algorithms and a set of binarization algorithms. At the stage of image processing, using multiple combinations instead of one specific combination of algorithms will help to achieve a more flexible and resistant operation of the recognition module to various types of defects in images. The developed model of the character recognition process will help to achieve a significant increase in the accuracy of recognition of unciphered documents due to the most efficient processing of images with different noise. In addition to removing noise, the improved model will allow to more accurately find the boundaries of images that have quality problems. The theoretical significance of the study is the result of modeling the character recognition process using many combinations of image processing algorithms to conduct a more extensive analysis and identify the most effective combinations of algorithms.