In this paper we propose a novel transform domain steganography technique—hiding a message in components of linear combination of high order eigenfaces vectors. By high order we mean eigenvectors responsible for dimensions with low amount of overall image variance, which are usually related to high-frequency parameters of image (details). The study found that when the method was trained on large enough data sets, image quality was nearly unaffected by modification of some linear combination coefficients used as PCA-based features. The proposed method is only limited to facial images, but in the era of overwhelming influence of social media, hundreds of thousands of selfies uploaded every day to social networks do not arouse any suspicion as a potential steganography communication channel. From our best knowledge there is no description of any popular steganography method that utilizes eigenfaces image domain. Due to this fact we have performed extensive evaluation of our method using at least 200,000 facial images for training and robustness evaluation of proposed approach. The obtained results are very promising. What is more, our numerical comparison with other state-of-the-art algorithms proved that eigenfaces-based steganography is among most robust methods against compression attack. The proposed research can be reproduced because we use publicly accessible data set and our implementation can be downloaded.
This is an open access article distributed under the Creative Commons Attribution License
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited