On robustness of camera identification algorithms
- PDF / 2,790,640 Bytes
- 22 Pages / 439.642 x 666.49 pts Page_size
- 13 Downloads / 208 Views
On robustness of camera identification algorithms Jarosław Bernacki1 Received: 27 August 2019 / Revised: 8 May 2020 / Accepted: 27 May 2020 / © The Author(s) 2020
Abstract In this paper we consider the problem of a privacy threat enabling tracing digital cameras by the analysis of pictures they produced. As thousands of images are processed at a mass scale, the threat may apply to most users of digital cameras. We consider a state-of-theart algorithm for digital camera identification proposed in Lucas et al. (IEEE Trans Inf Forensics Secur 1(2):205–214, 2006) and discuss strategies that can be used to bypass it, in order to make information about the camera unavailable. It turns out that many natural strategies like Gaussian blur, adding artificial noise or removing pixels’ least significant bit from the image does not prevent the identification of a camera unless a huge loss of image details is suffered. On the other hand, we show a method to bypass the camera identification with a just marginally more complex, yet not intuitive, method namely cropping the image on the edges and resizing to the original size using Lanczos resampling. Keywords Privacy · Hardwaremetry · Camera recognition · Camera fingerprint · Lanczos resampling
1 Introduction Since digital cameras are easily accessible and relatively cheap, many people take dozens of pictures on a daily basis. Furthermore, the pictures are shared using social media services or photo portals with friends, but also with thousands of coincidental onlookers. Posting hundreds of photos made using the same digital camera or a smartphone allows us to be linked to a certain device or a model, hence has negative consequences from the privacy point of view. The more privacy-aware of the social media users remove the meta-data indicating the time, location and the device model using image editors, however the linking a picture with a device does not require the additional meta-data. This can expose users’ Jarosław Bernacki
[email protected] 1
Cze¸stochowa University of Technology, al. Armii Krajowej 36, 42-200, Cze¸stochowa, Poland
Multimedia Tools and Applications
privacy to a serious threat if they post an image outside of their personalized social media profiles, identifying pictures’ author may lead from slight inconvenience like personalized spam to serious criminal activity like stalking or user personalized phishing. Naturally, the techniques of linking a device to an user might serve as well crime prevention. One of the most popular issue in image processing and image forensics is recognizing the camera based on images and using it as a “digital fingerprint” or proof of presence. An algorithm proposed by Luk´as et al.’s in [21] is considered to be most efficient and is commonly used by black- and whitehats alike. Many recent approaches using deep learning models use image denoising formula presented in this approach. The assumption of Luk´as et al.’s algorithm is to calculate the so called Photo-Response Nonuniformity Noise (PRNU) for each photo.
Data Loading...
