Security for Visual Data in Wavelet Representation

We propose a project in the area of multimedia security. Currently, wavelet-based representations of visual data are highly relevant due to their widespread use in compression standards (JPEG2000 and parts of MPEG-4) and proprietory multimedia applications. We focus on two main research directions: image/video encryption and watermarking. In the area of encrypting visual data, most current approaches concentrate on selective encryption of MPEG streams whereas in the watermarking area many approaches suffer from a lack of integrated security features (as opposed to robustness). For these reasons, we propose to investigate the impact of exploiting wavelet-domain specific features and properties within multimedia security applications in order to overcome the aformentioned shortcomings.

Multimedia security is a recent field important in the area of digital rights protection, digital rights management, and many more. Main emphasis of this project has been devoted to wavelet-based techniques in this area, technologies covered are watermarking, low cost encryption of visual data, and robust visual hashing. Wavelet transforms offer a certain degree of freedom in the choice of the actual specific transform in a decomposition. The main approach followed in this project is to exploit this degree of freedom to implement security techniques based on the idea of "secret transform domains" or "key dependent transform domains". In particular, this idea has been used as follows: 1) Watermarking: the actual wavelet transform domain used for watermark embedding is kept secret and controlled by a key. We were able to demonstrate a significant security benefit for a certain class of watermarking algorithms using this technique, where unauthorized watermark detection and unauthorized watermark removal attacks could be repelled. 2) Encryption: the actual wavelet transform domain used within a compression codec (like JPEG2000) is kept secret and controlled by a key. We have designed and evaluated lightweight encryption schemes with extremely low encryption effort, suited for low power mobile devices like PDAs or cell phones - this is achieved by restricting encryption to the parameters defining the transform, the data itself can be left unencrypted. The security of this approach turns out to be highly dependent on the compression scheme used. 3) Robust Visual Hashing: the actual wavelet transform domain used within a hashing scheme is kept secret and controlled by a key. We have demonstrated higher security for certain authentication focussed hashing techniques, whereas the even higher robustness of CBIR hashing schemes seems to inhibit an increase in security. Apart from the topics described, many other issues have been covered, e.g. bitstream based JPEG2000 encryption, multiple watermarking, permutation within wavelet compression pipelines, encryption of scalable MPEG and progressive JPEG data, and many more.