Using neural networks to recognize text labels of raster navigation maps.

The solution of the problem of text recognition on raster navigation maps is considered. The implementation of image processing by the method of colorimetric conversion to grayscale based on the Otzu algorithm for text recognition on navigation maps is described. The input data are files containing training data for the neural network (a set of images of individual symbols and a text document containing training data labels) and files of raster images of navigation maps without tilt. The output data is a vector image containing the results of recognition of text labels and a text file containing all recognized labels. The experimental part of the work was carried out for three types of neural networks (three-layer, five-layer and six-layer) trained on three data sets. The results of the study allow us to conclude about the combination of the number of neural network layers used and the number of data types in the training set. [ABSTRACT FROM AUTHOR]