Steganalysis by Subtractive Pixel Adjacency Matrix

By Tomás Pevný, Patrick Bas, Jessica J. Fridrich. IEEE Transactions on Information Forensics and Security, 2009

1,066 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/3c3e20b7114c85dc2413c407338c82218d671103

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Recasting Residual-based Local Descriptors as Convolutional Neural Networks: an Application to Image Forgery Detection (2017, methodology): “To associate a residual-based feature to an image, or an image block, the following processing chain has been successfully used in steganalysis [11, 21]:”
Ensemble Classifiers for Steganalysis of Digital Media (2012, background): “The SPAM set for the secondorder Markov model of pixel differences has a dimensionality of 686 [39].”
Steganalysis of LSB matching using differences between nonadjacent pixels (2016, methodology): “ALE [30] SPAM [24] DHCF COMs#1 DHCF COMs#2 SPAM and DHCF COMs#2”
Detecting GAN generated Fake Images using Co-occurrence Matrices (2019, methodology): “Other techniques involve calculating image residuals or passing the image through different ﬁlters before computing the co-occurrence matrix [7, 8, 9].”
Detecting CNN-Generated Facial Images in Real-World Scenarios (2020, background): “Lastly, several works use co-occurrence matrices to focus on irregularities in pixel-patterns, for example in steganalysis [12, 14, 38, 45, 46] and detection of forged images [11, 12].”
Universal distortion function for steganography in an arbitrary domain (2014, background): “For example, HUGO [15] defines the distortion as a weighted norm between higher-order statistics of pixel differences in cover and stego images [16], with high weights assigned to well-populated bins”
Autoencoder with recurrent neural networks for video forgery detection (2017, methodology): “Here, we use features inspired by work in ste-ganalysis [33, 34].”
Digital image steganography using universal distortion (2013, methodology): “HUGO [28] embeds by minimizing an embedding distortion defined as a weighted norm between the features of the cover and stego image in the SPAM feature space [27].”
Rich Models for Steganalysis of Digital Images (2012, methodology): “Personal use of this material is permitted.”
Deep Residual Network for Steganalysis of Digital Images (2019, background): “[1], [2] and Farid and Siwei [18] in early 2000’s and was later greatly improved by representing images with higher-order statistics of noise residuals or DCT coefficients [43], [49], [68].”
Minimizing Additive Distortion in Steganography Using Syndrome-Trellis Codes (2011, methodology): “Moreover, recent results from steganalysis i n both spatial [39] and DCT domains [40] suggest that the secure payload for digital image steganography is always fa r below 1/2.”
"Break Our Steganographic System": The Ins and Outs of Organizing BOSS (2011, methodology): “The image model was largely inspired by the Subtractive Pixel Adjacency Matrix (SPAM) steganalytic features [11], but steps have been taken to avoid over-ﬁtting to a particular feature set [9].”
An Overview on Image Forensics (2013, methodology): “To cope with �PE� postcompression, they presented another �rst-order difference-based detector which utilized the subtractive pixel adjacency matrix (SPAM) features [120].”
Steganalysis of least significant bit matching using multi-order differences (2014, methodology): “Table II shows that, in the detection of LSB matching, the performance of MODF is almost the same as that of SPAM and better than ALE and RDHF.”
Information Forensics: An Overview of the First Decade (2013, methodology): “To overcome this, subtractive pixel adjacency matrix (SPAM) features developed to perform steganalysis [43] can be employed to capture properties of an image’s conditional pixel difference distributions”
A Deep Learning Approach to Universal Image Manipulation Detection Using a New Convolutional Layer (2016, methodology): “These tools from steganalysis operate by building local models of pixel dependencies by analyzing the joint distribution of pixel value prediction errors, then extracting detection features from these joint distributions [16, 4].”