Computer scientists have developed a new tool for connecting digital media with their manufacturer – ScienceDaily

In a project aimed at developing intelligent tools to combat child exploitation, computer scientists at the University of Groningen have developed a system for analyzing noise produced by individual cameras. This information can be used to connect video or image to a specific camera. The results were published in journals SN Computer Science on June 4, 2022 and Expert systems with applications on June 10, 2022

The Netherlands is the main distributor of digital content showing child sexual abuse, as reported by the Internet Watch Foundation in 2019. To combat this type of abuse, forensic digital content analysis tools are needed to identify which images or videos contain suspicious children abusive content. Another unused source of information is noise in images or video frames. As part of an EU project, computer scientists from the University of Groningen, along with colleagues from the University of Leon (Spain), have found a way to extract and classify noise from an image or video that reveals the camera’s “fingerprint”. .

Bullet

“You can compare it to the specific channels of a bullet fired,” said George Azopardi, an assistant professor in the information systems research group at the Bernoulli Institute for Mathematics, Computer Science and Artificial Intelligence at the University of Groningen. Each firearm produces a specific pattern on the bullet, so forensic scientists can compare a bullet found at a crime scene with a specific firearm or connect two bullets found at different crime scenes with the same weapon.

“Each camera has some imperfections in its built-in sensors, which appear as image noise in all frames, but are invisible to the naked eye,” explains Azopardi. This creates camera-specific noise. Guru Bennabhaktula, a PhD student at both Groningen and the University of Leon, has developed a system for extracting and analyzing this noise. “In image recognition, classifiers are used to retrieve information about the shapes and textures of objects in the image to identify a scene,” says Benabhaktula. “We used these classifiers to extract camera-specific noise.”

Law enforcement

He created a computational model for extracting camera noise from video footage taken with 28 different cameras taken from VISION’s publicly available dataset, and used it to train a convolutional neural network. He then tests whether the trained system can recognize footage taken by the same camera. “It turns out that we can do this with 72 percent accuracy,” says Benabhaktula. He also notes that noise can be unique for a brand of camera, for a specific type and for individual cameras. In another set of experiments, he achieved 99 percent accuracy in classifying 18 camera models using images from a publicly available data set in Dresden.

His work is part of an EU project, 4NSEEK, in which scientists and law enforcement agencies are collaborating to develop smart tools to help combat child exploitation. Azopardi: “Each group was responsible for developing a specific forensic tool.” The model created by Bennabhaktula may have such practical use. “If the police find a camera of a child abuse suspect, they can link it to images or videos found on storage devices.”

Challenges

The model is scalable, adds Benabhaktula. “Using only five random frames of video, it is possible to classify five videos per second. The classifier used in the model has been used by others to distinguish over 10,000 different classes for other computer vision applications. This means that the classifier can compare the noise of tens of thousands of cameras. The 4NSEEK project has been completed, but Azzopardi is still in contact with forensic and law enforcement experts to continue this research line. “And we’re also working on identifying the similarity of the source between a pair of images, which has different challenges. This will form our next document on this topic.

Source of history:

Materials provided by University of Groningen. Note: Content can be edited for style and length.

Leave a Comment

Your email address will not be published.