Identity Document Recognition: Neural Network Approach

Abstract

The passport of the Russian Federation is the main document of Russian citizens. A Russian passport is required for the vast majority of transactions in banks, government agencies, education, etc. This paper presents a neural network approach to data recognition from the Russian Federation passport. The use of two neural networks is proposed. The first neural network based on SOTA neural network YOLOv5 to determine the location of areas with text. The second convolutional neural network with the topology chosen by the authors to recognize the text in the previously selected blocks. A console application was implemented to demonstrate the approach. The accuracy of the first network is 78.6%, the second is 97.4%.