Lossless Information Hiding Based On Image Features

Due to the rapid development of Internet and digital multimedia, an increasing number of digital products are published on the Internet in the form of images, audios and videos. These digital works can be obtained easily and conveniently, but their copyright protection and information security are greatly challenged. For this purpose, lossless information hiding(LIH) has been proposed as an effective solution, which embeds watermarks into published digital products based on the perception of human visual systems and the redundancy of digital media. LIH can not only protect copyrights of digital products by detecting watermarks, but also completely recover host media in lossless environment, which is widely used in medical, remote sensing and military fields. So far, a lot of LIH methods have been proposed; however, how to design novel watermark embedding and extraction models to improve the invisibility, capacity and robustness is of great importance. Therefore, this paper focuses on the challenging problems of LIH and the main contents are as follows.(1) In the existed LIH methods, most of them design the watermark embedding and extraction based on whole images without considering the local features of images, which cannot be effectively used to protect the information security of regions of interest in medical and remote sensing images. Therefore, this paper first utilizes feature detectors to detect feature points in host images and construct the local feature regions centered with the feature points. Secondly, the sub-images are designed to embed watermarks losslessly to remain the corresponding relationships of local feature regions during the watermark extraction procedure. In this method, the capacity can be controlled by adjusting the block size and scale factors, which can satisfy the requirements of different applications.(2) In order to improve the robustness of LIH methods, this paper proposes the nonnegative matrix factorization based robust lossless information hiding method by using Slantlet transform. In comparison with wavelet transforms, Slantlet transform has good characteristics, which plays a key role in improving the robustness. Firstly, the host image is transformed with Slantlet transform to obtain its low frequency coefficients. Secondly, the 1D coefficients vector is created by the non-negative matrix factorization. Thereafter, the frequency distribution histogram is built based on the frequency distribution and the polarity vector is constructed to generate the keys. The experimental results show that the proposed method achieves good invisibility and robustness.In summary, this paper combines the merits of image features and lossless information hiding and proposes two novel LIH methods, which can not only recover the watermarks completely but also improve the capacity, invisibility and robustness.