Research On Non-intrusive Digital Camera Image Source Identification

With the development of digital information technology, more users prefer to use electronic imaging device to take pictures and videos, digital images have been the important media and widely integrated into many areas. It will cause great social problems if the images involving copyright dispute, digital forensic and some criminal cases, the judiciary be an urgent need for images tracking and identification of sources. This paper will study the non-invasive technology for digital image source identification, using images information without human embed information, and this technology is essential to justice, images media and many other aspects of multimedia network. In this paper, combining the theoretical analysis and experimental verification to research on non-intrusive images source identification based on camera, the main contents and contributions are organized as follows:Firstly, the paper provided an overview of the relative background about non-intrusive digital images source identification, the concepts and current research status at home and abroad about digital image forensics. Then generalized and classified the existing relative research literature, discussed the facing problems, theoretical model and system framework on current images source identification.Secondly, the imaging process and working principle of general digital camera are presented, then analyzed the embedded noise in camera images information while imaging sensor taking images. Then generalized the current digital cameras source identification methods based on SPN, a non-invasive identification scheme is proposed for digital image source using correlation matching principle. It contains the extraction of features, correlation matching, testing, validation, recognition and classification, providing the auxiliary evidence for justice.Thirdly, in order to reduce the interference of scene stains and improve the accuracy of recognition rate, a novel filtering algorithm is proposed on extracting of SPN based on dual domain. The method use the wavelet transform and local adaptive edge-preserving bilateral filtering for SPN extraction, and the residual image matrix after filtering is the approximate SPN, then the classifier based on maximum correlation principle are constructed to identify the SPN has been extracted, and compared the results of SPN extraction approach using different color channels and luminance component of images. The experiments indicate that the proposed method effectively reduces the interference due to scene stains, a satisfactory accuracy has been achieved when the size of detected images are smaller.Fourthly, due to the limitation of existing methods, the SPN extracted in images are heavily contaminated by the scene details, and misidentification rate is high unless large size images are tested. In this paper, observed the risks with bayesian estimation based on minimum error probability for large size images. To facilitate image transmission and storage, most of the camera output images saved as JPEG format even after double JPEG compression by consumer. This paper discussed the impact of SPN under JPEG compression and correction operations, results show that the proposed method remains at a better stability and robustness for JPEG compression, and improves the practical application potential on images source identification.Finally,we summarized our research work for the thesis,and discussed some research topics and directions relative to this work in the future.