Digital Camera Identification

Forensically linking an image to a particular camera can be very useful. In a court of law, the origin of a particular photo may be crucial evidence in a case against child pornographers or industrial espionage. ere are a number of approaches that can be taken, but most have severe issues. Metadata in the header may be lost if the image is saved into a different format and can be easily forged. Police needing to verify the authenticity of their images can use special cameras that embed fragile watermarks or store a hash of the image on a secure memory card, but in most situations the image will have been taken on a normal consumer camera. What’s needed then is a way to link the image data itself to the camera. In this essay I will be examining the method proposed by Lukáš et al. to identify cameras by their unique sensor pattern noise [5]. Previous approaches include an investigation into the use of supervised learning based on a vector of numerical features extracted from the spatial and wavelet domains [4]. However, this only achieved 95% accuracy even in the best case; not good enough for most forensic work. Another research project looked at identifying cameras by pixel defects [2] but these are oen eliminated in the post-processing of modern cameras and as there are some cameras without any defects in the sensor, this is again suboptimal.