A trigger-perceivable backdoor attack framework driven by image steganography

Weixuan Tang, Jiahao Li, Yuan Rao, Zhili Zhou, Fei Peng

Abstract: Highlights•The core of launching effective backdoor attack is enabling the victim model to perceive the trigger for making wrong predictions.•Formulate trigger injection and perception under backdoor attack into message hiding and extraction under steganography.•Owing to the optimization paradigm of steganography and the reversibility of INN, trigger can be injected in a model-perceivable manner.•The transformation layer can simulate the processing-based backdoor defense, and increase the robustness of backdoor attack.