Double JPEG compression forensics based on a convolutional neural network
Wang and Zhang EURASIP Journal on Information Security
Double JPEG compression forensics based on a convolutional neural network
Qing Wang 0 1
Rong Zhang 0 1
0 Key Laboratory of Electromagnetic Space Information, Chinese Academy of Sciences , Huangshan Road, 230027 Hefei , China
1 Department of Electronic Engineering and Information Science, University of Science and Technology of China , Huangshan Road, Hefei , China
Double JPEG compression detection has received considerable attention in blind image forensics. However, only few techniques can provide automatic localization. To address this challenge, this paper proposes a double JPEG compression detection algorithm based on a convolutional neural network (CNN). The CNN is designed to classify histograms of discrete cosine transform (DCT) coefficients, which differ between single-compressed areas (tampered areas) and double-compressed areas (untampered areas). The localization result is obtained according to the classification results. Experimental results show that the proposed algorithm performs well in double JPEG compression detection and forgery localization, especially when the first compression quality factor is higher than the second.
Blind image forensics; Double JPEG compression; Convolutional neural network; Classification
1 Introduction
Generally, blind forensics techniques utilize statistical
and geometrical features, interpolation effects, or feature
inconsistencies to verify the authenticity of image/videos
when no prior knowledge of the original sources is
available. Because JPEG compression may cover certain
traces of digital tampering, many techniques are effective
only on uncompressed images. However, most
multimedia capture devices and post-processing software
suites such as Photoshop, output images in the JPEG
format, and most digital images on the internet are also
JPEG images. Hence, developing blind forensics
techniques that are robust to JPEG compression is vital.
Tampering with JPEG images often involves
recompression, i.e., resaving the forged image in the JPEG format
with a different compression quality factor after digital
tampering, which may introduce evidence of double
JPEG compression. Recently, many successful double
JPEG compression detection algorithms have been
proposed. Lukáš and Fridrich [1] and Popescu and Farid [2]
have performed some pioneering work. They analyzed
the double quantization (DQ) effect before and after
tampering and found that the discrete cosine transform
(DCT) coefficient’s histograms for an image region that
has been quantized twice generally show a periodicity,
differing from the DCT coefficient’s histograms for a
single-quantized region. Chen and Hsu [3] identified
periodic compression artifacts in DCT coefficients in
either the spatial or Fourier domain, which can detect
both block-aligned and nonaligned double JPEG
compression. Fu et al. [4] and Li et al. [5] reported that DCT
coefficients of single-compressed images generally follow
Benford’s law, whereas those of double-compressed
images violate it. In [5], they detect double-compressed
JPEG images by using mode-based first digit features
combined with Fisher linear discriminant (FLD) analysis.
Fridrich et al. [6] applied double JPEG compression in
steganography. The feature they used is derived from the
statistics of low-frequency DCT coefficients, and it is
effective not only for normal forged images but also for
images processed using steganographic algorithms.
However, a commonality among all algorithms
discussed above is that they estimate only the compression
history of an image, which cannot indicate exactly which
region has been manipulated. In fact, the localization of
tampered regions is a basic necessity for meaningful
image forgery detection. Nevertheless, to the best of our
knowledge, only few forensics algorithms can achieve it.
Lin et al.’s algorithm [7] was the first to automatically
locate local tampered areas by analyzing the DQ effects
hidden among the DCT coefficient’s histograms. The
authors applied the Bayesian approach to estimate the
© 2016 The Author(s). Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and
reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to
the Creative Commons license, and indicate if changes were made.
probabilities of individual 8 × 8 block being untampered.
In this way, the obtained block posterior probability map
(BPPM) would show a visual difference between
tampered (single-compressed) regions and unchanged
(double-compressed) regions. To locate the tampered
regions more accurately, Wang et al. [8] utilized the prior
knowledge that a tampered region should be smooth
and clustered and minimized a defined energy function
using the graph cut algorithm to lo (...truncated)