In the last few years, human recognition systems based on machine learning, specifically on deep neural networks, are on the rise. This, in turn, increased researchers' interest in protection against such systems. In this paper, we propose a method of designing camouflage patterns that will protect a person from being recognized by a human observer and recognition systems based on deep neural networks. This type of camouflage is based on adversarial examples generated by DCGAN. The paper describes experiments on protecting a person from being recognized by Faster-RCNN Inception V2 and Faster-RCNN ResNet101. The results of such experiments have demonstrated the efficiency of the proposed method in cyberspace where we can access every pixel fed to the image recognition system.
Volume 11 | 02-Special Issue
Pages: 506-516