Bio inspired optimization for universal spatial image steganalysis

https://doi.org/10.1016/j.jocs.2017.06.014Get rights and content

Highlights

Abstract

Universal Image steganalysis is a two class optimization problem. This research uses S-UNIWARD spatial steganographic algorithm to create stego images from 500 cover images. The image features extracted in spatial domain are noise models of neighbouring pixels giving 1000 × 34671 features. Ant Lion Optimization (ALO) is used to get best image features (1000 × 381 features). The classifiers used are Single (SVM and MLP) and Fusion classifiers (Bayes, Decision Template, Dempster Schafer). All fusion classifiers and SVM give classification accuracy of 99.3%. Thus Fusion classifiers with ALO act as best universal steganalyser in spatial domain.

Introduction

Steganography is the art of concealed communication where the very existence of the secret data in a digital media is not noticeable. Among the different steganographic carrier media, image is found to be more secure as the human eye is less perceptible to minor changes in image pixel values. Apart from the popular image steganographic tools over internet [30] many researcher have developed their own steganographic algorithms in both JPEG and spatial domains of images. Owing to a wide variety of steganographic schemes universal (blind) method of steganalysis has become the choice of steganalysers [26]. Universal image steganalysis intends to distinguish the clean (cover) image from the embedded (stego) image and hence is a two class optimized classification problem. Modelling an image for steganalysis enables easy representation of the image in terms of specific parameters. The higher the model of the image, the more accurate is the classification as there are many parameters to represent the image. Generally this accuracy is obtained at the expense of dimensionality of the feature space. The higher dimension of the feature space pose problem of computational complexity in terms of time and space [4], [5]. The curse of dimensionality can be overcome by proper optimization of the model parameters.

Existing optimization techniques include Principle Component Analysis (PCA), Factor Analysis [24], [17], Independent Component Analysis (ICA) [2], Self Organising Maps (SOM) [31], [23], Principal Factor Analysis, Maximum likelihood factor analysis [21], scalable feature extraction techniques [12], [37], Linear Discriminant Analysis (LDA) [3], Expectation Maximization [7] and clustering based Vector Quantization (VQ) [23]. The main disadvantage of these statistical dimension reduction techniques is their convergence to local minima leading to inappropriate feature selection (reduced feature set). To overcome this, bio inspired algorithms use natural selection principles of Darwinian theory of evolution. These nature inspired methods tend to optimize the feature set by selecting the fittest data variable according to optimization function. These optimization techniques depend on the stochastic (metaheuristic) behaviour of the data variables and are classified as GA based and swarm based techniques. Few recent swarm based techniques include Particle Swarm optimization (PSO), Harmony search (HS) algorithm [41], Artificial Bee Colony (ABC) developed by Karaboga [8], the firefly algorithm of Yang [39], Cuckoo search algorithm [40], Ant Colony Optimization (ACO) [1]. The firefly algorithm uses the flashing pattern to attract potential prey (select features). PSO algorithms provide solutions in terms of current position and best possible position. In Artificial Bee Colony (ABC) algorithm the selection of food source is correlated with memorization of the new position. In this research, the most recent doodlebug based optimization proposed by Mirjalili [29] has been modified and applied for universal spatial image steganalysis. The Ant Lion Optimization algorithm depicts the hunting behaviour of antlions in their larvae stage. They build cone shaped traps and wait at the bottom of the cone for prey (ants). When an ant comes inside the cone, the antlions throw sand towards the edges of the cone which makes the ants to slide down to the bottom. When an ant (prey) slides and reaches the bottom of the cone, it is consumed by the antlion. The antlion then rebuilds the cone and waits for another ant (prey). The positions of the ants are synonymous to the image features in the search space and the best features are synonymous to the position of antlions.

Section snippets

Spatial steganography and steganalysis

As steganographic embedding can be done in either spatial or JPEG domain of image, steganalysis needs to extract the corresponding feature in either of the domain. Current spatial steganographic algorithms like Least Significant Bit Matching (LSBM), Edge Adaptive (EA) algorithm, Highly Undetectable Steganography (HUGO) [38], S-UNIWARD [35] tend to embed the secret information in the noisy regions of the images as modelling these noisy regions is difficult during steganalysis. HUGO embedding

Experimental investigation

The images for this research have been taken from the BOSS (Break Our Steganographic System) data base [25]. These are full resolution raw images in .pgm format acquired from five different cameras (Leica, Nikon, Panasonic, Canon EOS, Pentax). We have chosen 500 images as cover images for steganographic embedding. This research has used the Spatial − Universal Wavelet Relative Distortion (S-UNIWARD) embedding algorithm [35] which is a modification of the Wavelet Obtained Weights (WOW).

Discussion and comparison with previous research

Based on the results illustrated above, it can be understood that ALO based optimization of image features leads to better classification accuracies irrespective of the type of classifier. It could be noted that SVM, Bayes, Decision template, Dempster Schafer classifiers have same classification accuracies (99.3%) and MLP is slightly lesser (98.6%). This proves that ALO gives best images features (381 features) from the extracted rich model of 34671 features. The state of the art approach by

Significance of ALO algorithm for universal image steganalysis

The inherent significances of ALO algorithm [29] as compared to the other nature inspired algorithms are

  • Guaranteed exploration of the search space (feature space) due to random walk of ants and random selection of ant lions.

  • Due to relocation of antlions to best position of ants, global minimal areas (best features) are preserved.

  • There a very few adjustable parameters in the ALO algorithm.

  • Universal image steganalysis by ALO method of feature selection proves to be significant due to the

Conclusion

Obtaining the optimal image features for accurate image steganalysis has been a challenge in universal steganalysis. This research has implemented Ant Lion Optimization, a bio-inspired meta-heuristic algorithm to obtain the best image features for spatial image steganalysis. The extracted image features include SPAM and Min-Max features from 106 noise sub models under six different categories. Each noise model has 625 elements and are formed from co-occurrence matrices of order 4. The total

J. Anita Christaline completed her B.E. Degree in Electronics and Communication Engineering in 1996 from Bharathidasan University, Trichy. She is recipient of All India Exam − GATE (Graduate Aptitude Test in Engineering) during 2003 and secured high score. She received her M.E. Degree in Applied Electronics in the year 2006 from Anna University. She is currently working as Assistant Professor in SRM University, Chennai where she is pursuing her Doctoral Degree. Her area of interest is Image

References (41)

  • S. Mirjalili
    (2015)
  • X.S. Yang et al.

    Cuckoo Search via Levy Flights, in Proc. Of World Congress on Nature & Biologically Inspired Computing (NaBic 2009)

    (2009)
  • A. Al-Ani

    Ant Colony Optimization for Feature Subset Selection

    (2005)
  • A. Hyvarinen

    Survey on independent component analysis

    Neural Comput. Surv.

    (1999)
  • A.M. Martinez et al.

    PCA versus LDA

    IEEE Trans. Pattern Anal. Mach. Intell.

    (2001)
  • J. Anita Christaline et al.

    Critical review of image steganalysis techniques

    Int. J. Adv. Intell. Paradigms

    (2015)
  • J. Anita Christaline et al.

    Optimized JPEG steganalysis

    Int. J. Multimedia Ubiquitous Eng.

    (2016)
  • J. Anita Christaline et al.

    Bio-Inspired computational algorithms for improved image steganalysis

    Indian J. Sci. Technol.

    (2016)
  • C.M. Bishop et al.
    (2012)
  • D. Karaboga

    An Idea on Honey Bee Swarm for Numerical Optimization, Technical Report

    (2005)
  • D. Zou et al.

    Steganalysis based on Markov model of thresholded prediction-error image

    Proceedings of IEEE, International Conference on Multimedia and Expo

    (2006)
  • G. Chen et al.

    On Optimal Feature Selection Using Harmony Search for Image Steganalysis

    (2012)
  • G. Gül et al.

    A new methodology in steganalysis: breaking highly undetactable steganograpy (HUGO)

  • H. Li et al.

    Efficient and Robust Feature Extraction by Maximum Margin Criterion, Proc. Conf. Advances in Neural Information Processing Systems

    (2004)
  • H. Sajedi

    Image steganalysis using artificial bee colony algorithm

    J. Exp. Theor. Artif. Intell.

    (2017)
  • H. Farid et al.

    Detecting hidden messages using higher order statistics and support vector machines

  • I. Avcibas et al.

    Image steganalysis with binary similarity measures

    EURASIP J. Appl. Signal Process.

    (2005)
  • I. Avcibas et al.

    Steganalysis using image quality metrics

  • I.T. Jolliffe

    Principal Component Analysis

    (1986)
  • J. Fridrich et al.

    Rich models for steganalysis of digital images

    IEEE TIFS

    (2011)
  • Cited by (0)

    J. Anita Christaline completed her B.E. Degree in Electronics and Communication Engineering in 1996 from Bharathidasan University, Trichy. She is recipient of All India Exam − GATE (Graduate Aptitude Test in Engineering) during 2003 and secured high score. She received her M.E. Degree in Applied Electronics in the year 2006 from Anna University. She is currently working as Assistant Professor in SRM University, Chennai where she is pursuing her Doctoral Degree. Her area of interest is Image Steganography and Steganalysis by computational intelligence. She has published many papers in this field. She is member IEEE and IET.

    R. Ramesh was born in Kanyakumari, India, 1976. He received B.E.Degree in Electronics and Communication Engineering in 1998 and M.E. degree in Communication Systems in 2000, both from Madurai Kamaraj University, India. He has been awarded Doctoral degree in SRM University in the year 2009 for his research work on Testing the Stability of two dimensional recursive filters. He is currently working as a Professor in the Department of Electronics and Communication Engineering at Saveetha Engineering College, Chennai. India.H is current research interests concern digital image and signal processing.

    C. Gomathy acquired a B.E.degree in Electronics and Communication Engineering from Government College of Engineering, Tirunelveli, in the year 1986, and an M.S. degree in Electronics and Control Engineering from BITS, Pilani, in 1992. She also obtained an M.S. degree from Anna University in 2001. She obtained her Ph.D. in the area of Mobile Ad hoc networks from College of Engineering, Anna University, Chennai, India, in the year 2007. She has published over 70 research papers in National and International conferences and journals. Her areas of interest include mobile ad hoc networks, high-speed networks, and wireless sensor networks.

    D.Vaishali was born in Pune, India, 1969. She received B.E .Degree in Electronics and Telecommunication Engineering in 1994 and the M.E. degree in Communication Systems in 2002, both from Pune University, India. She is a research scholar in the SRM University and her research work is progressing in the field of Image Processing with wavelet Transforms. She is currently working as a Asst. Professor in the Department of Electronics and Communication Engineering at SRM University, Chennai. India.S he is member IEEE and IET.

    View full text