Image provenance inference through content-based device fingerprint analysis

The last few decades have witnessed the increasing popularity of low-cost and highquality digital imaging devices ranging from digital cameras to cellphones with builtin cameras, which makes the acquisition of digital images become easier than ever before. Meanwhile, the ever-increasing convenience of image acquisition has bred the pervasiveness of powerful image editing tools, which allow even unskilled persons to easily manipulate digital images. As a consequence, the credibility of digital images has been questioned and challenged. Under the circumstance where digital images serve as the critical evidence, e.g., presented as evidence in courts, being able to infer the provenance of an image becomes essential for recovering truth and ensuring justice. As an important branch of digital forensics, image provenance inference aims to determine the original source of a digital image. The provenance of an image provides forensic investigators with rich information about the originality and integrity of the image. It does not only look for answers to the question of which device has been used to acquire a given image, but also conveys other implications of the credibility of an image. For example, the inconsistent provenance information from different regions of an image indicates that the image may have been tampered with. This chapter mainly introduces and discusses several intrinsic device fingerprints and their applications in image provenance inference. These fingerprints arise from either the hardware or software processing components in the image acquisition pipeline and exhibit themselves as specific patterns or traces in the image. Analyses of these fingerprints provide useful information for inferring the image provenance and uncovering underlying facts about the image. In the remainder of this chapter, we will first discuss why the techniques that based on digital watermark and metadata are impractical or unreliable for image provenance inference in Section 9.2 and 9.3, respectively. In Section 9.4, we will introduce several intrinsic device fingerprints arising from different processing components in the imaging pipeline of a device. In Section 9.5, we will concentrate on Sensor Pattern Noise (SPN) and discuss in detail its applications in image prove-