@techreport{oai:ipsj.ixsq.nii.ac.jp:00232625, author = {イ, チャンソク and 岡田, 啓 and 和田, 忠浩 and ベンナイラ, シャドリア and 片山, 正昭 and Changseok, Lee and Hiraku, Okada and Tadahiro, Wada and Chedlia, Ben Naila and Masaaki, Katayama}, issue = {21}, month = {Feb}, note = {人が視認しにくいディスプレイーカメラ可視光通信は画像や動画などの視覚情報に視認できないような方法でデータを埋め込み,視覚情報と同時にデータを送る手法である.その一つとして,単眼深層学習モデルの敵対的サンプルを用いたディスプレイーカメラ可視光通信を提案した.しかし,敵対的サンプルは小さい画素値の変化のため,移動,回転,拡大,ぼかし,ノイズ,色変化などの通信路による影響を受けやすい.そこで,本研究ではその対策となる Expectation over Transformation (EOT) を導入する.EOT は予想できる通信路の影響を敵対的サンプルの生成過程に加えることで,敵対的サンプルの通信路の影響への耐性をつける手法である.本研究では EOT 導入の効果を実装実験により評価し,この新たな手法の有効性を示す., Hidden display-camera visible light communication is a method of embedding data in visual information such as images and videos in a way that is not visually noticeable, and transmitting the data simultaneously with visual information. As one of these methods, display camera visible light communication using adversarial examples on a monocular deep estimation model was proposed. However, adversarial samples are susceptible to the effects of communication channel noise such as movement, rotation, enlargement, blurring, noise, and color changes, as they involve small pixel value changes. Therefore, in this study, we introduce Expectation over Transformation (EOT) as a countermeasure. EOT is a method that adds expected communication channel noise to the generation process of adversarial samples to impart resistance to communication channel noise. In this study, we evaluate the impact of introducing EOT through experiments, demonstrating the effectiveness of this novel approach.}, title = {単眼深度推定モデルの敵対的サンプルを用いたディスプレイーカメラ可視光通信の評価実験}, year = {2024} }