ALGORITHM FOR RECOGNITION HIGHLY CORRUPTED QR-CODES

Abstract

The intensive development of information technology has led to the creation of data exchange systems, that using combined compression, protection against damage and information storage. Such systems usually use matrix codes, that allow to store a large amount of information compactly and recognized it quickly by scanning equipment. In this article has been solved the task of recognition and correction of high damaged matrix codes, namely QR codes, where there is a high level of noise, there are no key elements or colors are overlaid. Such images are not recognized by decoding software as the structure of the detectors of elements is damaged and the correcting ability of the built-in Reed-Solomon codes doesn't allow to correct the necessary part of errors. That's why the algorithm based on the usage of artificial Hamming-Lippmann neural network with the base of samples and image processing in the sliding window mode is offered, which simplifies the learning process of the network without appliance of labor-intensive computational operations, large volumes of memory and high time consumption, even for images with high resolution and big size. The network learning process consists of two parts: image processing,            sample recognition and correction. In order to achieve correct recognition, it is necessary to identify experimentally the optimal parameters of learning, thanks to which the matrix of samples the rows will differ sufficiently from each other. For this purpose, the authors developed software in C# language, with the helping of which the necessary experimental researches were performed. The conditions of correct work of the neural network (optimal values of the sliding window size and the threshold for different sizes of QR-code images) has been determined, as well as cases, when errors of recognition and instabilities of its outputs are possible. The results of the researches show, that the developed algorithm can be applied as an additional procedure of recognition and correction of QR codes in different data exchange systems.