A Data Set of Handwritten Characters of the Russian Language

The availability of open data sets is an important factor for the development of machine learning. There are not many data sets that contain national handwritten characters published today. The paper provides an overview of the published data sets and their comparative analysis. As can be seen from the review, there are very few Russian-speaking data sets in the presented ones, and most of them are a set of images of handwritten or printed characters. The article presents a data set collected by the authors describing handwritten symbols of the Russian language. This data set contains character writing samples from 12 writers and contains uppercase and lowercase characters, as well as numbers – a total of 76 character classes. The data set additionally includes some spelling patterns of whole words. The information about the symbols are presented in the form of coordinates of the trajectories of the writer's pen during the process of drawing the symbol. There is also a program code similar to the EMNIST data set that allows you to convert character trajectories into images. There is also a program code that allows you to convert character trajectories into images, similar to the EMNIST data set. The paper describes the methodology of data collection and processing. In addition, the recognition accuracy of the basic LeNet-5 model of the presented data set is compared with the EMNIST data set, which is similar in number of classes. For the presented data set, the quality is shown to be slightly worse than for EMNIST. However, the presented data set, at the same time, has important properties – since the data is set not only by the image matrix form, but also by the trajectory of movement – this allows you to take into account the peculiarities of pen movement, which can be used not only for the purpose of classifying characters, but also for the purpose of recognizing personal features of handwriting, transferring writing style, graphological analysis.