Moderate embed cross validated and feature reduced Steganalysis using principal component analysis in spatial and transform domain with Support Vector Machine and Support Vector Machine-Particle Swarm Optimization

The fast evolution of Information and Digital technology had given way for internet to be an effective medium for communication. This has also paved way for data exploitation. Therefore, users must protect their data from misuse. This led to the emergence of security framework like Information Hiding. Steganography and Steganalysis are of the two primary techniques in the field of Information Hiding. Steganography is the science of concealing confidential information, while steganalysis is the art of detecting the existence of that information. The primary goal of this research is to address the general concept of steganalysis, and various breaches associated with it. It involves a blind statistical steganalysis technique that is led in Joint Photographic Experts Group (JPEG) text embedded images by extracting features that illustrate an alteration during an embedding. The images used as embedding medium are uncalibrated and the percentage of the embedding used in this study is 50%. The text embedding is done using various steganographic schemes in the spatial and transform domain. The steganographic schemes considered are Least Significant Bit (LSB) Matching, Least Significant Bit (LSB) Replacement, Pixel Value Differencing and F5. After steganographic embedding of the data, the first order, second order, extended Discrete Cosine Transform (DCT) and Markov features are extracted. Then, Principal Component Analysis (PCA) is used as a system for feature dimensionality reduction. Furthermore, the technique of machine learning is incorporated by means of a classifier to identify the stego image and cover image. Support Vector Machine (SVM) and Support Vector Machine with Particle Swarm Optimization (SVM-PSO) are the classifiers examined in this paper for a comparative study. Moreover, the concept of cross-validation is also incorporated in this work. Six dissimilar kernel functions and four diverse samplings are used during classification to check on the effectiveness of the kernels and sampling in classification.