Digital forensics has taken a step forward, thanks to a technique developed at State University of New York at Binghamton. A group of researchers led by Jessica Fridrich, associate professor in the department of electrical and computer engineering, has applied for two US patents on a method that links a digital camera to the images it produces, which may enable law enforcement officials to prove connections between suspects and evidence. Approaches to camera identification have included examining the dead pixels in images and looking at the artifacts left by processing. Camera manufacturers share algorithms and processing, however, so they could pinpoint a number of cameras but not an individual one. The challenge was to come up with a signature that is distinctive to a sensor and not to a camera brand. The new technique rests on the fact that every original digital picture is overlaid by a weak noiselike pattern of pixel-to-pixel nonuniformity unique to the camera, the result of inhomogeneities in the silicon wafer used to make the image sensor and of the manufacturing process. The noise is the same in all images taken with the camera. The pattern of noise extracted from a digital image can be used to identify the particular camera that produced it — in this case, a Canon PowerShot G2. Courtesy of Jessica Fridrich.The pattern is extracted by examining the digital images on a computer using a de-noising filter, similar to the way that digital watermarks are detected. Fridrich’s team has found that having one image is sufficient to identify the pattern and to match it to those in other images. Pattern identifies cameraThis can be used in child pornography cases, for example, when a camera has been confiscated. Images on a suspect’s computer may be analyzed to determine whether they were taken with that camera. The method also may determine the presence or absence of the expected pattern, which Fridrich noted could be used to substantiate or question the reliability of a digital photo offered as evidence. With millions of pixels in an average digital photo, the number of variations is extremely high, she said, so the likelihood that two cameras share the same signature is extremely low. However, she estimated that the only way to prove this is to analyze millions of cameras. The next step for the researchers is further testing to improve the reliability and speed of the technique. They also want to apply it to images from scanners and digital video cameras, which pose additional challenges and opportunities.