Steganalysis of QIM Steganography

This paper proposes a statistical steganalysis method for quantization index modulation (QIM) based steganography. We have shown that, in general, plain-quantization (quantization without message embedding) reduces local-randomness (or increases local-correlation) in the resulting quantized-object and QIM-stego exhibits higher level of local-randomness than the corresponding quantized cover. The local-randomness of the testimage is used to capture traces left behind by quantization (with or without message embedding). We model the distortion due to quantization as a gamma distribution. The parameters of this gamma distribution are estimated using maximum likelihood estimators. Distributions of the parameters estimated from the quantized-cover and the QIM-stego images are used to develop a generalized likelihood ratio test (GLRT) to distinguish between the cover and the stego images. Effectiveness of the proposed method is evaluated using a large set (over 35000 images) consisting of test-images obtained using sequential as well as random message embedding. Experimental results show that the proposed scheme can successfully detect QIM-stego images with very low false rates (Pfp < 0.015 and Pfn < 0.03) for sequential embedding and false rates (Pfn < 0.038 and Pfp < 0.013) for random embedding. In addition, performance comparison with existing state of the art also shows that the proposed method performs significantly superior than the selected methods.