Identification of image global processing operator chain based on feature decoupling

Abstract

Image authenticity verification is an important issue to be studied, which has attracted growing attention recently. Most of the existing forensic methods are aimed at detecting a specific manipulation. However, due to the superimposed processing artifacts caused by using different operations to forge images, the image global processing operator chain identification, which is composed of multiple global manipulations in a certain order, remains a challenge. In this paper, we focus on detecting multiple manipulations and identifying the order of these manipulations. By analyzing the relationship between blind signal separation and operator chain identification, we find that the independent source features of different operations will be coupled when the image is processed by multiple operations, which is similar to what in blind signal separation. Therefore, it is reasonable to formulate the problem of operator chain identification with blind signal separation. Then, a feature decoupling method is proposed to estimate the source feature from the coupled features by optimizing a decoupling matrix. These estimated decoupled features are valid evidence for operator chain identification. For the realistic scenario where images are saved in JPEG format, the comparison with some state-of-the-art methods demonstrates that the proposed method could identify operator chains with better performance.