Digital image source shooting device identification method based on hardware fingerprint correlation

Abstract

The invention discloses a digital image source shooting device identification method based on hardware fingerprint correlation. The method comprises the steps of firstly extracting a camera fingerprint of a to-be-discriminated camera, then calculating a set of correlation description values of pictures belonging to or not belonging to the camera and the camera fingerprint, finally performing hypothesis testing on the to-be-detected picture, and determining whether the to-be-detected picture belongs to the camera or not. According to the invention, a source camera identification problem is converted into a hypothesis testing problem, so that the identification accuracy is effectively improved under the condition of increasing a small amount of computing and storage resources.

Description

technical field [0001] The invention belongs to the field of digital passive evidence collection, and relates to a method for identifying digital image source shooting equipment based on hardware fingerprint correlation. Background technique [0002] With the development of science and technology and the rapid popularization and wide application of various smart phones, digital cameras and other mobile devices, video and images have become important carriers of information like text and voice, and have been integrated into daily work, study and life. It is an effective tool for conveying information and proving objective facts. However, frequent incidents such as the dissemination of illegal content, disputes over the copyright of photographic works, digital forensics in court, and falsification of news and scientific research images have brought challenges to the credibility of images and videos, as well as the tracing of the source of illegal content. Therefore, it is of ...

AI Technical Summary

Problems solved by technology

At present, there are a variety of extraction algorithms for PRNU. The basic idea is to collect images taken by different devices, use the PRNU extraction algorithm to extract image fingerprints from these images, and then estimate the device fingerprints by calculating the mean value or maximum likelihood estimation. , and then calculate the correlation between the fingerprints of each device and the given test image, so as to determine the camera object that took the given test image. The traditional threshold discrimination method only uses a single correlation feature to judge and the support vector machine, convolutional neural network Machine learning methods such as machine learning have the disadvantages of slow computing speed, large resource requirements, and complex models.

Images