China
Africa
Explore chapters and articles related to this topic
Part Review on Multimedia Security
Published in Ling Guan, Yifeng He, Sun-Yuan Kung, Multimedia Image and Video Processing, 2012
Alex C. Kot, Huijuan Yang, Hong Cao
Photo-response nonuniformity (PRNU) noise is the dominant noise of the sensor pattern noise, which is caused by the imperfect manufacturing process. Lukas et al. [144] proposed to extract PRNU noise pattern from normal photos for individual camera identification. The reference PRNU pattern is learned through averaging synchronized training PRNU patterns from a camera, which suppresses the random noise and enhances the pattern noise. With 9 cameras and 300 training photos per camera, the results show that the PRNU patterns can be used as features to identify individual cameras based on a correlation detector with a close-to-zero false rejection rate when the false acceptance rate is fixed at 0.1%. Backed with the good identification results, Lukas et al. [162] further extended the PRNU pattern approach to discover the local tampered region of interest (ROI). By correlating the local PRNU patterns extracted from a test image with the synchronized reference PRNU patterns based on different sliding block shapes and sizes, the forged ROI is automatically determined and it shows relative reliable identification at a JPEG quality factor of 70. Chen et al. [150] extended this PRNU approach by improving the preprocessing techniques, the noise model, and the correlation detector. As a result, fewer training photos are needed to learn the reference PRNU pattern and better results are achieved. Based on a sliding block size of 128 × 128, the correlation statistics for each pixel is measured and converted into a probabilistic score of tampering, which is subsequently used to determine whether a pixel is tampered.
Photo forgery detection using RGB color model permutations
Published in The Imaging Science Journal, 2022
Most of the time, fake photos are created for fun and rumor. However, fake photos may be the reason for problems in political, social, and legal systems. Though, there is a requirement for photo’s authenticity without any availability of a digital signature or digital watermark on it. In most cases, the source of the photo is unknown. In this paper, the emphasis is on blind or passive photo forgery detection. In passive detection, no source and historic information about the photo is available. The only photo itself without any prior information is available to check its originality. Broadly, copy-move forgery and splicing forgery types of fake photos are popular. Fake photos are created using multiple photos in the splicing forgery. In the copy-move forgery, the photo self-portion is copied and pasted into another region to hide or duplicate some information. The detection of fake photos is becoming challenging due to the evolution of artificial intelligence and deep learning [1,2] based methods. Splicing forgery and copy-move forgery are detected using Camera fingerprint, lighting conditions disorder, color illumination inconsistency, JPEG compression artifacts, and statistical anomaly. The camera fingerprints like Photo Response Non-Uniformity (PRNU) noise and Color Filter Array (CFA) are used in the photo forgery detection. Nevertheless, these approaches are more suitable for splicing detection and localization of foreign portions in the spliced image. In [3], PRNU analysis is performed to locate the small size patches of forgery. Multiple tampering probability maps at different scales are combined for reliable results. Chierchia et al. [4] applied the Bayesian estimation with Markov random field for source identification. PRNU is applied for forgery detection. The papers [5,6] inform that a single light sensor is generally used in capturing devices to save costs. However, CFA is used to overcome this issue and generates three colors in the RGB color photo. The CFA is implemented with some interpolation method that leaves some fingerprints in each photo. This fingerprint helps in forgery detection. The illumination-based method [7] gives a good result on IFS-TC 2012 challenge dataset [8]. A physics-based method is used in which integral is performed over a gradient field of the subject to find the direction of incident light. This method claims robustness against color variation and works on the whole object area, unlikely other methods. In [9,10] methods, a 3-D lighting model is discovered to detect photo forgery. The 3-D lighting model highlights the lighting inconsistency in a fake photo. These methods claim their robustness against photo resizing and JPEG compression also. Some methods [11,12] rely on JPEG compression artifacts, i.e. blocking and quantization. Though, these methods are bound to a particular type of photo format.