Error Correction of Sample Pair Analysis Based on Support Vector Regression

Abstract

For the typical quantitative steganalysis method -- sample pair analysis, a new method is proposed to correct the error of it based on support vector regression. The proposed method uses the support vector regression to fit the function of estimation error with respect to the embedding ratio, local variance and saturation. Then for the given image, the estimated embedding ratio of sample pair analysis, local variance and saturation pixel ratio are fed to the trained model to estimate the error and correct the estimated embedding ratio. Experimental results show that the proposed method can reduce the standard deviation of the estimation errors by larger than 0.0024 for embedding ratio less than 1.0 and reduce the mean of the estimation errors by larger than 0.0074 for embedding ratio less than 0.6.