The Development of Methods and Algorithms for Text Recognition Using Neural Network

The main purpose of this work is to increase the accuracy of classification in the task of text recognition. The object of development is the recognition algorithm, which includes two separate modules: 1. Image processing module; 2. Recognition module. Also, there were developed new methods necessary to obtain a better recognition accuracy. During the development of methods and algorithms for text recognition, the following results were obtained: 1. The average accuracy on the test data has reached 92%. For real data, the accuracy varies in the range of 80-90%; 2. The algorithm is stable to rotations of the whole image, and to rotations of individual symbols. Also, the algorithm is resistant to some printing defects; 3. Partially solved the problem associated with the recognition of similar pairs of characters. The results obtained can be useful both in research and in the development of such systems as the recognition of automobile numbers, extensions of ancient texts and the automation of processes related to the processing of textual information (passport control). Also, some of the parts of the study can be useful in medicine (image processing). The work consists of 91 pages and contains 57 figures, 5 tables and 49 sources.