Image Souce Identification Based On Camera Artifacts Detection

Digital forensics is a new research area for against recent e-crime. Multi-media forensics is one of the hottest research directions in this field. There have many subjects in multi-media forensics. Among these subjects, digital image forensics, which aims the truth, integrity, and reliability detection of digital image contents are gotten more attention. However, traditional digital image forensics (active forensics) only works if there has authentication pre-processing before image publication. Considering the difficult of spread, the use of traditional way is limited. For requiring needs of image source truth and reliability's authentication, digital image passive forensics was born. There are two hot branches of digital image passive forensics, truth detection and source identification. For need of market, digital image source identification of camera-oriented becomes popular. Its major research directions include camera model identification, camera identification, and the distinguishing of photo-graphic image and computer-graphic image. Digital camera imaging leaves its 'special artifacts'on final image. These'artifacts'are too tiny to be seem by human's eyes. But theoretically, it can be detected by image analysis technology. Most recent solutions of digital image source identification of camera-oriented were based on camera artifacts detection.Digital image source identification has a short history. Research work still has lots of shortage. Common problems include large feature dimension, unsatisfactory detection accuracy, etc. Facing the increase of image source device types and image processing software, it is better to meet the accuracy and scalability of detection by researching on detection method with less feature dimension and high accuracy. It is significance for accelerating practical of digital image source identification.This paper firstly introduces current digital image source identification, and then discusses three technlogies, camera model identification based on CFA and interpolation, camera model identification based on overall imaging, the distinguishing of photographic and computergraphic image based on multi image components,. Better performance image source identification methods are designed. Besides, our work is expected to meet the requirement of verifying reliability of digital image source in criminal investigation, which has high value for both study and practial. Main work in this paper is as follow.(1) Camera model identification based on CFA and interpolation artifacts detectionTraditional camera model identification based on CFA and interpolation artifacts detection has a large feature dimension. The reason causes this problem is:Traditional way usually estimates demosaicing by its approximate interpolation filter under different CFA hypothesis. This will lead to large feature dimension. Aim to these problems, we do research on the relationship of raw and interpolation color components, propose a camera model identification based on CFA and interpolation. The difference between interpolations is captured by statistical difference between raw and interpolation color components, instead of approximate interpolation filter coefficients. The feature dimension is decreased with close detection accuracy. Besides, interpolation filter estimation is no need in ours method, which decrease running-time effectively. New method is helpful for practical of camera model detection based on CFA and interpolation.(2) Camera model identification based on multi-step Markov modelIn traditional camera model identifications based on overall imaging, one-step Markov based method has unsatisfactory identification accuracy. The reason is: Traditional way just uses correlation between one-step neighbor DCT coefficients for modeling, does not dig out more. Aim to this problem, we work on a multi-step Markov model; point out it can dig out more correlation of DCT coefficients, increase detection accuracy. We use multi-step Markov model instead of one-step Markov model, and use averaging algorithm to decrease feature dimension. A camera model identification based on multi-step Markov model is designed. The detection accuracy is increased with slight higher feature extraction running time and lower feature dimension. The positive effection of better modeling is discussed and proven.(3) The distinguishing of photo-graphic image and computer-graphic image based on multi imaging componentsTraditional distinguishing of photo-graphic image and computer-graphic image has unsatisfactory detection accuracy. The reason is:it always distinguishes the difference between details of nature image and non-nature image by wavelet methods. However, wavelet bases are not best bases for presenting line singularity of image edge and details, which cause unsatisfactory accuracy. Aim to this problem, optical imaging can be presented by white balance features. Combining white balance feature set with a CFA feature set and a sensor noise feature set, our distinguishing method can better reflect camera imaging existing or not. The experiment shows that our method can get high distinguishing accuracy with lower feature dimension. It proves white balance detection can works on the distinguishing of photo-graphic and computer-graphic images.To sum up, our research work analysis some shortage of current digital image source identification based on camera artifacts detection. Aim to these problems, we exploit better performance camera artifacts detection and how to use it on camera model identification, the distinguishing of photo-graphic and computer-graphic images. Our work is helpful for the research of digital image source identification, both for theoretical and application aspects. The whole research work is concluded at last part. So is future work.