Research On Image Operator Chain Forensics Based On Feature Coupling Correlations

With the widespread of various digital media recording devices,digital images have become an important carrier for transferring and obtaining information in people’s daily lives.At the same time,the rapid development of powerful and easy-to-use media editing tools has made the modification and editing of digital images more and more convenient.Therefore,in the face of the increasingly serious image tampering situation,it is urgent to develop corresponding forensics technology to verify the authenticity and integrity of digital images.Passive image forensics technology analyzes the inherent characteristics of the image to realize the authenticity and integrity of digital images.As an emerging branch in the field of image forensics,forensics of the image operator chain is one of the significant research contents in the field of information security.The forensics of image operator chain aims to completely reveal the types of editing processing operations and order of operations that digital images may have experienced.In the actual detection process,the image tampering operation chain is also regarded as a composite of a series of basic operations.This paper analyzes the coupling effect between different operations from the feature level and proposes a forensics framework for image operator chain based on the correlation of feature coupling.Then,the research on the operation type detection and order discrimination of the image operator chain was conducted separately.The research results are as follows:(1)A general forensics framework for actual image operator chain forensics research is designed.To detect the types of tampering operations that may be included in the image operator chain,first,formulate the detectability problem of specific operations in the multiple operator chains into a multi-hypothesis testing problem.By measuring the correlation relationships between the real hypothesis and the detection hypothesis with the information entropy,give the theoretical quantification standard for the detectability of one specific operation in the multi-operator chains.After determining the types of operations included in the image operator chains,in order to further determine their topological order,the hypothesis test theory is used to analyze the distinguishable problem of the topological order of the operator chain.Via calculating the conditional transfer possibility between the real hypothesis and the detected hypothesis,display the criteria decisions for distinguishing the topological order of operations.Then,by analyzing the naturally existing correlation between the characteristics of the tampering operation,the Pearson correlation coefficient is used to mine the coupling relationship between different basic tampering operation types from the feature level and design the coupled features.Finally,a multi-class feature extractor is used to classify the coupled feature content,and a general theoretical framework that can be used for image operator chain forensics is designed.(2)The original digital image is usually stored in a JPEG format before being subjected to a tampering operation,which is called a Pre-JPEG image.Considering that the blocking effect in JPEG compression will change the content of the tampering feature,in the feature extraction stage for the forensics of the Pre-JPEG image operation chain,we redesigned the anti-JPEG compression coupled feature by analyzing the forensic technology of different single operations in JPEG formats.Besides,unlike a single classification model,we have constructed a multi-classification model that can identify complex operations to capture and learn fingerprint features of different tampering operations.The experimental results show that the coupling features we constructed can effectively achieve the complex operation type detection and order discrimination of PreJPEG images under different quantization coefficients;the designed multi-classification model can deeply mine the coupling features of tampering operations,thereby making the operation chain forensics framework more robust and efficient.