Variable characteristics based blind detection of hidden information
Kang Sun · Journal of Zhejiang University(Engineering Science) · 2007
To detect the stego images generated with smaller embedded information and steganographic schemes of Steghide and MB2,a statistical model was proposed based on functions of joint photographic coding experts group(JPEG) image and its approximate version image.Variable characteristics were captured in discrete cosine transform(DCT) domain of JPEG images and from their decompressed versions with these functions.In DCT domain,statistics of DCT alternating current coefficients were extracted from variable characteristics.In spatial domain,statistics of pixel value differences distributed at two sides of DCT blocks were captured in decompressed images.A multi-dimensional feature vector was computed from the statistics of an original JPEG image and the calibrated one,and then a multi-dimensional statistical model was built.Detecting results for eight steganographic schemes show that a blindly detecting algorithm based on the presented model and support vector machine provides good detecting effect,and that compared with traditional algorithms,the algorithm shows higher detecting rate and lower false positive.