Abstract: The trustable security design based on RO（Ring Oscillator） is an effective way to check whether the hardware Trojan is inserted. The reason is that the insertion of hardware Trojan will introduce low temperature areas and cause the RO thermal profile change. Considering the weak thermal signal of hardware Trojan, this paper proposes a detail enhancement method that utilizes both the time dimension and space dimension. Firstly, the adaptive filtering is used to reduce the noise in the time dimension. Next, the thermal map is sharpened by the detail enhancement method based on the guided filtering. The experimental results show that the SNR of processed thermal map is 20.1104 dB, which is 6.1421 dB more than that of original thermal maps, 3.5956 dB more than that of temporal de-noising only approach, and 3.6050 dB more than that of spatial detail enhancement only approach. The proposed approach performs better de-nosing effectiveness and richer details. The insertion of hardware Trojan can be detected through verifying the change of the low-temperature regions profiles.

Key words: hardware Trojan detection, thermal image enhancement, adaptive filtering, guided filtering

摘要： 将RO环（Ring Oscillator，环形振荡器）作为可信任安全性设计是一种检测硬件木马是否植入的有效方式，硬件木马的植入会形成低温区，从而导致RO环热边界改变。面对低温区红外信号相对较弱的特点，提出了一种融合了时间维和空间维的细节增强方法，即先在时间维使用自适应滤波降低噪声，然后在空间维使用基于引导滤波的细节增强方法锐化图像。实验结果表明，经此方法处理后红外图像信噪比为20.110 4 dB，相比原始红外图片提高了6.142 1 dB，相比单纯降噪提高了3.595 6 dB，相比单纯细节增强提高了3.605 0 dB，具有更优秀的降噪性能和更丰富的细节，可以显化低温区热边界并检测硬件木马的植入。

关键词: 硬件木马检测, 红外图像增强, 自适应滤波, 引导滤波