| Authors |
C.S. Lee, H. Okada, T. Wada, C. Ben Naila, M. Katayama |
| Title |
Validation of Hidden Screen-Camera Communication Systems Using Adversarial Examples on CNN Depth Estimation Model through Simulation |
| Authority |
International Conference on Materials and Systems for Sustainability (ICMaSS), A6-III-5 |
| Summary |
Screen-camera communication is widely recognized for its low cost and intuitiveness. Moreover, hidden screen-camera communication enables simultaneous transmission of visual information, such as images, and data. However, it requires a certain level of communication performance and visual quality. In this study, we verify screen-camera communication using adversarial examples of a deep learning model. Adversarial examples can alter the output of the target model with slight changes of the input so that it can be used to embed data in screen-camera communication, specifically in the output of a deep learning depth estimation model. Additionally, we introduce the Expectation over Transformation (EOT) technique to enhance the robustness of adversarial examples to noise, which is their weakness. We validate the effectiveness of this communication method in a noisy environment through communication channel simulations in this study and manage to achieve a bit error rate of 0.035 or less even in simulations with noise. |
| 年月 |
2023年12月 |
| DOI/Handle |
|
| 開催場所 |
Nagoya, Japan |
| 研究テーマ |
可視光通信／光無線通信
機械学習
|
| 言語 |
英語 |
| 原稿/プレゼン資料 |
 /  (ローカル限定) |
