Information System for Analyzing the Quality of Object Recognition in Pictures

Abstract

The work aims to study the quality of image recognition of objects by neural networks. An analysis of the popular neural networks, which are publicly available, was carried out. The developed information system allows you to interact with several neural networks conveniently and apply filters to the image. Using the developed program a series of experiments was conducted to determine the impact of image transformations on the quality of their recognition. The results of the work can be used to audit the current capabilities of neural networks for recognition of the CAPTCHA and can be used as a basis for developing neural network-resistant graphical protection methods.