JPEG Universal Steganalysis Based on Symmetric α-stable Model
WU Ming-qian · Jisuanji gongcheng · 2014
Aiming at the problem of JPEG steganography detection, a universal steganalysis approach is proposed based on DCT coefficient distribution model. A better bi-parameter Symmetric α-stable(SαS) model is built according to the statistical regular of the DCT coefficient distribution. The model parameters, estimated by solving a discrete function parameter optimization problem using advanced genetic algorithm, are employed to be the steganalytic features and calibrated to improve its sensibility. A feature-matched one-class classifier with linear computation complexity is designed. Experimental results show that the method is reliable in steganalysis for all kinds of JPEG steganography. When the embedding rate is 25%, its mean detection rate achieves 76.1%, which is at least 5.5% higher than the traditional model-based steganalysis methods, it has high detection performance.