Improving blind image steganalyzer performance using third order spam features and random rotation forest ensemble mpsvm

Abstract

In recent years, complex steganography techniques have been used to hide the secret messages in an innocuous looking digital cover medium. Besides its importance, nowadays this technique has been used for antisocial activities such as international attacks, stealing the military secrets, confidential trading/ technical messages, hacking, electronic payments, etc. Researchers are working on blind/universal steganalysis system which helps to detect the presence of hidden message in an innocuous-looking cover medium. Though the steganography techniques have got more development, many existing blind steganalyser result in less detection accuracy. The key functions of the steganalyser are feature extraction and classification. The focus of this present research is to develop novel steganalyser with high accuracy. The framework of the proposed work comprises four phases namely, stego image set generation, image denoising, feature extraction and classification. In the stego image set generation phase, the stego images are generated in three ways: (1) Existing tools: Digimarc, PGS (Pretty Good Signature), S-Tools, Steganos, Jsteg (2) Existing algorithms: Spread spectrum Method, Wavelet based data hiding, YASS (Yet Another Steganography Scheme) and (3) the proposed algorithm. Some of the existing approaches have poor payload capacity and poor image quality (visual discrepancy). As a result an improved steganography algorithm, is proposed which uses a modified quantization table to embed the secret bits in the two LSB (Least Significant Bits) of the middle frequencies. The importance of denoising phase is to eliminate Gaussian noise from the image and to obtain noise residuals. newline