Suppressing Random Artifacts in Reference Sensor Pattern Noise via Decorrelation

Abstract

Photo response nonuniformity noise is a component of sensor pattern noise (SPN) of digital cameras. It characterizes the imaging devices of cameras and has been widely used in source camera identification (SCI). However, the reference SPN always be contaminated by unwanted nonunique artifacts. Previous works reported that these nonunique artifacts were periodic signals shared among cameras, which undergo the same or similar in-camera processing procedures. In this letter, we first illustrate that these nonunique artifacts, in addition to periodic signals, also contain random signals caused by the imperfection of SPN extraction method. To improve the quality of SPN and the performance of SCI, then we propose a scheme to suppress these random signals by decorrelation. In our proposed method, the reference SPN is modeled as a white Gaussian noise, while the random signals introduced by SPN extraction method are correlated noises, we suppress these correlated components to purify the SPN and enhance the performance of SCI. The experimental results on “Dresden” database show that the proposed scheme achieves better receiver operating characteristic curves than the state-of-the-art methods.