关键词: 通用隐写分析, 支持向量回归, 损失函数, 核函数

Abstract: In order to solve the problem that the majority of general steganalysis methods cannot estimate the secret message length, this paper proposes an improved general quantitative steg-analysis method that can estimate secret message length. 132 dimensional features describing the correlations between DCT coefficients are extracted from stego images. Support vector regression is used to learn the mapping between feature vectors and the relative embedding change rates and construct steganalyzer model. Embedding rates are estimated through new feature sets and steganalyzer model. Simulation is performed on stego images embedded with F5 , MB and outguess steganographic algorithms. The results of simulation reveal that the proposed quantitative steganalysis is feasible to estimate the embedding ratio of stego images in practice.

Key words: quantitative steganalysis, support vector regression, loss function, kernel function