Significant method for the recognition of structured documents with manuscript filling on the basis of a convolutional neural network

Abstract

The methods of image processing used in computer-readable handwriting recognition systems are considered and investigated. Two main tasks are solved: the definition of the structure of the document and the recognition of the handwritten characters. For character recognition, the convolutional neural network is used. It is proposed to use an array of informative signs of a symbol to improve its learning efficiency. Estimates of efficiency indicators are presented.