A Features Decoupling Method for Multiple Manipulations Identification in Image Operation Chains

摘要

Recently, many forensic techniques have been developed to detect the use of a certain processing operation. When utilizing several manipulations to alter an image, artifacts left by manipulations that have been applied later can potentially disguise traces left by manipulations that were applied earlier. Therefore, the detection of manipulations become difficult. In this paper, we focus on identifying the manipulations in an image operation chain composed of multiple manipulations in a certain order. To address this issue, we analyze the relationship between manipulations identification and blind signal separation. Then, we propose a features decoupling method based on blind signal separation, which decouples the coupled features due to the superimposed processing artifacts and exploits the decoupled features to identify multiple operations. The experiments carried out on two image operation chains confirm the effectiveness of the proposed method.