Breaking ALASKA: Color Separation for Steganalysis in JPEG Domain

This paper describes the architecture and training of detectors developed for the ALASKA steganalysis challenge. For each quality factor in the range 60-98, several multi-class tile detectors implemented as SRNets were trained on various combinations of three input channels: luminance and two chrominance channels. To accept images of arbitrary size, the detector for each quality factor was a multi-class multi-layered perceptron trained on features extracted by the tile detectors. For quality 99 and 100, a new "reverse JPEG compatibility attack" was developed and also implemented using the SRNet via the tile detector. Throughout the paper, we explain various improvements we discovered during the course of the competition and discuss the challenges we encountered and trade offs that had to be adopted in order to build a detector capable of detecting steganographic content in a stego source of great diversity.

[1]  Patrick Bas,et al.  An Empirical Study of Steganography and Steganalysis of Color Images in the JPEG Domain , 2018, IWDW.

[2]  Jessica Fridrich,et al.  Effect of JPEG Quality on Steganographic Security , 2019, IH&MMSec.

[3]  Jimmy Ba,et al.  Adam: A Method for Stochastic Optimization , 2014, ICLR.

[4]  Jessica J. Fridrich,et al.  Feature-Based Steganalysis for JPEG Images and Its Implications for Future Design of Steganographic Schemes , 2004, Information Hiding.

[5]  Tomás Pevný,et al.  Is ensemble classifier needed for steganalysis in high-dimensional feature spaces? , 2015, 2015 IEEE International Workshop on Information Forensics and Security (WIFS).

[6]  Jessica J. Fridrich,et al.  Steganalyzing Images of Arbitrary Size with CNNs , 2018, Media Watermarking, Security, and Forensics.

[7]  Jiwu Huang,et al.  Large-Scale JPEG Image Steganalysis Using Hybrid Deep-Learning Framework , 2016, IEEE Transactions on Information Forensics and Security.

[8]  Yi Zhang,et al.  Steganalysis of Adaptive JPEG Steganography Using 2D Gabor Filters , 2015, IH&MMSec.

[9]  Jessica J. Fridrich,et al.  Low-Complexity Features for JPEG Steganalysis Using Undecimated DCT , 2015, IEEE Transactions on Information Forensics and Security.

[10]  Jiangqun Ni,et al.  An efficient JPEG steganographic scheme based on the block entropy of DCT coefficients , 2012, 2012 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).

[11]  Mo Chen,et al.  Deep Residual Network for Steganalysis of Digital Images , 2019, IEEE Transactions on Information Forensics and Security.

[12]  Jessica J. Fridrich,et al.  Steganalysis Features for Content-Adaptive JPEG Steganography , 2016, IEEE Transactions on Information Forensics and Security.

[13]  Andrew D. Ker.  On the Relationship Between Embedding Costs and Steganographic Capacity , 2018, IH&MMSec.

[14]  Kilian Q. Weinberger,et al.  On Calibration of Modern Neural Networks , 2017, ICML.

[15]  Guanshuo Xu,et al.  Deep Convolutional Neural Network to Detect J-UNIWARD , 2017, IH&MMSec.

[16]  Yun Q. Shi,et al.  Uniform Embedding for Efficient JPEG Steganography , 2014, IEEE Transactions on Information Forensics and Security.

[17]  Patrick Bas,et al.  The ALASKA Steganalysis Challenge: A First Step Towards Steganalysis , 2019, IH&MMSec.

[18]  Bin Li,et al.  WISERNet: Wider Separate-Then-Reunion Network for Steganalysis of Color Images , 2018, IEEE Transactions on Information Forensics and Security.

[19]  Andrew D. Ker.  The Square Root Law of Steganography: Bringing Theory Closer to Practice , 2017, IH&MMSec.

[20]  Bin Li,et al.  Large-scale JPEG steganalysis using hybrid deep-learning framework , 2016, ArXiv.

[21]  Tomás Pevný,et al.  The square root law of steganographic capacity , 2008, MM&Sec '08.

[22]  Jason Weston,et al.  Curriculum learning , 2009, ICML '09.

[23]  D. Sculley,et al.  TensorFlow Estimators: Managing Simplicity vs. Flexibility in High-Level Machine Learning Frameworks , 2017, KDD.