Метод оптичного розпізнавання текстів на основі неорієнтованих псевдографів
Вантажиться...
Файли
Дата
2018
Автори
Буряк, В.Г.
Буряк, В.В.
Buriak, V.
Buriak, V.
Назва журналу
Номер ISSN
Назва тому
Видавець
Хмельницький національний університет
Анотація
В статті описується розроблений метод оптичного розпізнавання текстів з використанням неорієнтованих псевдографів як спосіб представлення зображень символів, що розпізнаються. Растрове зображення векторизується, і на основі аналітичного представлення векторного зображення символу із застосуванням визначених правил формується псевдограф. Розпізнавання символу, представленого псевдо графом, здійснюється за допомогою нейронної мережі. Застосування розробленого методу дозволяє підвищити точність розпізнавання.
The article describes a developed optical character recognition method, which uses undirected pseudographs as a way of representing images of characters that are recognized. A bitmap image is vectorized, and on the basis of the analytic representation of the vector image of the character using definite rules a pseudograph is formed. Recognition of the character, which is represented as a pseudograph, is performed using the neural network. Application of the developed method allows to increase the accuracy of recognition. The method of optical recognition of texts with presentation of symbols in the form of non-oriented pseudographs, which are constructed on the basis of isolated attributes of the analytical representation of vector image of recognized symbols, is proposed. The application of the developed method provides an opportunity to reduce the amount of data that needs to be processed, compared with approaches based on raster image recognition. This is done by representing symbols in the form of non-oriented pseudographs that contain information about the topological characteristics of the symbols, namely: the orientation of the lines in the space and the order of connecting them points. In contrast to the direct use of the sequence of commands and the points of the analytical representation of vector graphics in the recognition subsystem, the method does not take into account the actual data on the location of the points (their coordinates) when recognizing them. The use of topological, rather than metric, characteristics allows us to increase the accuracy of the recognition of texts whose images are obtained in conditions of noise and optical distortions. Using the method allows you to improve the quality of character recognition from images obtained in unfavorable conditions. This system can recognize characters of texts with an accuracy of 97.99%.
The article describes a developed optical character recognition method, which uses undirected pseudographs as a way of representing images of characters that are recognized. A bitmap image is vectorized, and on the basis of the analytic representation of the vector image of the character using definite rules a pseudograph is formed. Recognition of the character, which is represented as a pseudograph, is performed using the neural network. Application of the developed method allows to increase the accuracy of recognition. The method of optical recognition of texts with presentation of symbols in the form of non-oriented pseudographs, which are constructed on the basis of isolated attributes of the analytical representation of vector image of recognized symbols, is proposed. The application of the developed method provides an opportunity to reduce the amount of data that needs to be processed, compared with approaches based on raster image recognition. This is done by representing symbols in the form of non-oriented pseudographs that contain information about the topological characteristics of the symbols, namely: the orientation of the lines in the space and the order of connecting them points. In contrast to the direct use of the sequence of commands and the points of the analytical representation of vector graphics in the recognition subsystem, the method does not take into account the actual data on the location of the points (their coordinates) when recognizing them. The use of topological, rather than metric, characteristics allows us to increase the accuracy of the recognition of texts whose images are obtained in conditions of noise and optical distortions. Using the method allows you to improve the quality of character recognition from images obtained in unfavorable conditions. This system can recognize characters of texts with an accuracy of 97.99%.
Опис
Ключові слова
оптичне розпізнавання текстів, граф, векторизація, растрове зображення, векторне зображення, нейронна мережа, optical character recognition, graph, vectorization, bitmap image, vector image, neural network
Бібліографічний опис
Буряк В. Г. Метод оптичного розпізнавання текстів на основі неорієнтованих псевдографів [Текст] / В. Г. Буряк, В. В. Буряк // Вимірювальна та обчислювальна техніка в технологічних процесах. – 2018. – № 2. – С. 99-106.