Source Camera Forensics Based On Pattern Noise

Digital image has become a major way for transmitting information with its immediate, convenient and easy handling characteristics in modern communications. In recent years, as a kind of multimedia information, digital image brought lots of convenience to society but at the same time provided tools for vicious acts, so its authenticity and reliability is constantly being questioned. To be a sort of evidence, digital image has a high probative force, and has played an increasingly important role in judicial system and the criminal investigation, therefore, the source camera identification technology which can ensure the reliability of image source is significant for maintaining social stability and justice.This paper focuses on the research of source camera forensics of digital images. To begin with the imaging process of digital camera, discussing the camera's software and hardware impact on image characteristics, analyzing the performance and summarizing the problem of the existing algorithms, then coming to a conclusion that the source camera forensics method based on camera hardware can better distinguish different cameras. Finally, proposed the method of this paper based on Pattern Noise which relative with camera sensor and lens to meet the requirements of forensic technology. The specific contributions of this paper are as follows:(1) Proposed a source camera identification method based on the correlation between principal component of Pattern NoiseFirstly, analyzing the composition of the extracted noise. There is no ideal denoising filter that can directly obtain the image's systematical noise now, the noise obtained through the filter is actually an estimate value which contains other random noise; at the same time, the pattern noise is a high frequency information, and the compression process for JPEG format images is carried out on its high frequency, this could reduce the Pattern Noise to a certain extent. In response to these problems, this paper proposed an effective way to obtain the principal component of Pattern Noise, so as to enlarge the validity of systemic noise in source camera forensics. Experiment results show that this method effectively improve the correlation of Pattern Noise between images from the same camera, also improve the low accuracy on distinguishing JPEG images. However, due to the different threshold of different camera, a given photo may be sentenced to different camera, so the method needs to be improved.(2) Proposed a source camera identification method based on Pattern Noise featuresIn order to improve the existing problems of the method above and low accuracy of the existing feature-based method for distinguishing cameras of the same brand, this paper considered the problem from the perspective of feature extraction, synthetically took spatial domain, frequency domain and wavelet domain characteristics into account, extracted 338 dimension features. Experiment results show that this is an effective method for source camera identification with less restrict, for existing samples, the average accuracy on distinguishing camera of the same brand reach 98.68%, has great superiority compared with the typical methods based on internal operation algorithm of the camera.