Personal Documents Identification System Development Using Neural Network

摘要

The problem of analyzing and classification of digital images, that contain the documents, is very common for finance and economic spheres, for example, for bank and government agencies. Analyzing the documents differs from photos classification since the documents very often have low contrast and are similar to each other. For the personal documents classification, the structure of the convolutional neural network has been designed and developed. To achieve high accuracy of the recognition, that is very important, working with documents, the optimal parameters for the designed network were determined experimentally. The problems, that occur using such system, are mentioned. A method, used to avoid overfitting of the network, is described. The experimental network has been approved by practical experiments, their results are presented.