Semantic-integrated software watermarking with tamper-proofing

The existing works of software watermarking have the intrinsic defects: watermarking is independent of program semantics and have weak strength and resilience to state-of-the-art reverse engineering such as symbolic execution, dynamic taint analysis and theorem proving. In this paper, we propose a semantic-integrated watermarking with tamper-proofing to mitigate such problems. This work chooses neural network as the “integrator” and skillfully integrates the watermarking and tamper-proofing module into program semantics. The difficult of reverse engineering or tampering with watermarked program is equal to extracting the rules from neural networks, which had be proven as a NP-hard problem. We have deployed our work in SPECint-2006 benchmarks to evaluate the overhead, strength and resilience. Experiment results show that our watermarking could effectively resist the state-of-the-art reverse engineering, and the introduced overhead is acceptable.