Passive blind image forensics based on the noise properties of digital camera

With the development of digital technology, the creation and manipulation of digital images become much easier, and the technologies used in doctoring become more and more hard to be detected. As a result, the unique stature of photographs as a definitive recording of events is being diminished. More and more people don’t trust what they see in the pictures, so image forensics is necessary to be studied. Today, digital images are widely used in the mass media, courts of law or military information. So the research on image forensics will be very important for these areas.

This thesis studies passive-blind image forensics, which is a method to detect tampered images without prior encryption information embedded. The approach of this thesis is based on the detection of the camera pattern noise, which serves as a unique identification fingerprint of a camera. This is achieved by averaging the noise obtained from multiple images using a denoising filter. Through studying the correlation between the noise residual of a whole image or the region of an image and the reference pattern of a camera, we could decide the origin of the image or whether there are forgeries in the image.

In this thesis, we first review the research area of passive-blind image forensic. Then we provide a detailed description of the theory, the realization, and the experiment results of our method. Furthermore, we point that our method has some limitations and what we could try to improve this method. Using some previous papers in this area as references, we improve an existed detection system.

Experiments on approximately 300 images taken with four consumer digital cameras are used to estimate false alarm rates and false rejection rates. Furthermore, this thesis applies the above method on the detection of forgeries in digital images. In individual regions in the image, the forged region is determined as the one that lacks the pattern noise. We propose automatic, manual, and semiautomatic methods to detect the copy-paste forgery.

Figure 1 (a) Original Image (b) Forgery Image

Figure 2 (a) Automatic Detection Result (b) Semiautomatic Detection Result

Figure 3 (a) Original Image (b) Noise Residual Image