Steganalysis using image quality metrics 论文

详细信息

发表期刊/会议

IEEE Transactions on Image Processing

发表日期

2003-02-01

发表年份

2003

摘要

We present techniques for steganalysis of images that have been potentially subjected to steganographic algorithms, both within the passive warden and active warden frameworks. Our hypothesis is that steganographic schemes leave statistical evidence that can be exploited for detection with the aid of image quality features and multivariate regression analysis. To this effect image quality metrics have been identified based on the analysis of variance (ANOVA) technique as feature sets to distinguish between cover-images and stego-images. The classifier between cover and stego-images is built using multivariate regression on the selected quality metrics and is trained based on an estimate of the original image. Simulation results with the chosen feature set and well-known watermarking and steganographic techniques indicate that our approach is able with reasonable accuracy to distinguish between cover and stego images.