Blind Quantitative Steganalysis using SVD Features

Abstract

Blind quantitative steganalysis form an important part of forensics tools. It unveils detail about hidden payload capacity without prior information about steganographic tool used for embedding. This paper proposes a novel approach for blind quantitative steganalysis using Singular Value Decomposition (SVD) features. SVD features are extracted from images, fed to regressor models for learning relationship between features set and embedded capacity in training phase. This relationship is used to estimate embedded payload in testing phase. Experimental study has been carried out in MATLAB. Statistical measures MAE and IQR are used to compare proposed quantitative steganalysis based on SVD features with quantitative steganalysis based on histogram features.