Scaling factor estimation on JPEG compressed images by cyclostationarity analysis

Scaling factor estimation is one of the most important topics in image forensics. The existing methods mainly employ the peak of the Fourier spectrum of the variance on image difference to detect the scaling factor. However, when the image is compressed, there will be additional stronger peaks which greatly affect the detection ability. In this paper, a novel method to estimate the scaling factor on JPEG compressed images in the presence of image scaling before the compression is proposed. We find the squared image difference can more effectively obtain the resampling characteristics, and we will mathematically show its periodicity. To further improve the detection ability, we analyze the flat block. It also produces periodic peaks in the spectrum, meanwhile which are enhanced by JPEG compression. To solve this problem, a method based on interpolation on the flat block is developed to remove these influences. The experimental results demonstrate that the proposed detection method outperforms some state-of-the-art methods.