الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 191 |  |  |
| | | from | 191 | | |
Abstract
Nowadays, it is so easy to maliciously manipulate multimedia data such as adding and removing objects. Therefore, image authentication becomes more important to assert the integrity of the various multimedia data forms. In this dissertation, authentication of digital images and video has been addressed. In the beginning, the conventional vector quantization (VQ) attack is modified to be capable of effectively attacking selffrecovery image authentication. Then, multiple description coding (MDC) is discussed and we propose to use a block-based MDC scheme that can be employed in selffrecovery techniques. In this MDC scheme, each block of an image is transformed to the frequency domain using the discrete cosine transform (DCT) and two groups are established from selected block coefficients. As a necessary condition, at least the DC coefficient must be in the two groups. Using a JPEG quantization table and a fixed bit allocation table, the coefficients of the two groups are quantized and encoded to generate two block descriptions.
Existing self-recovery authentication techniques have been reported as vulnerable to various attacks with insufficient security. Therefore, a new fragile image authenntication technique with self-recovery is proposed using MDC. In this technique, the image of interest is divided into blocks. For each block, a cryptographic signature (i.e., hash or message digest) and two description codes are computed and then inserted into the least significant bits (LSBs) of two relatively distant blocks. Two schemes are presented to choose the distant blocks. Also, a block signature copy is inserted into the block itself. The LSBs of each block are encrypted for more security. Embeddding block signature copies into two distant blocks makes a doubly linked chain that thwarts the VQ attack. One description is enough to recover the missing block with an acceptable quality, using MDC enhances the reliability of recovering/reconstructing the altered blocks.
Moreover, secure self-recovery authentication is presented using blocks of variableesizes that effectively prevent the VQ attack from constructing the codebook. Three partitioning schemes may be used to split an image into variable-size blocks: 22level tree partitioning, multi-level asymmetric binary tree partitioning and multi-level asymmetric quad tree partitioning. The cryptographic block signature is computed ilnrl t]lPn thp hlnrk j<; rp<;izpn to rnmnlltp its tWO nescriptions llsing ]\t(DC’. A hlock
signature copy is embedded into the block itself, and two block signatures with two block descriptions are inserted into the LSBs of two relatively-distant blocks. To choose the distant blocks, we propose a new scheme to grantee that the chosen blocks are relatively far from the considered block. Using a cryptosystem, the LSBs of each block are encrypted.
A new image hashing scheme is also presented using the discrete wavelet transsform (DWT) and non-negative matrix factorization (NMF). To survive JPEG commpression and additive noise, a semi-fragile authentication technique is proposed using the proposed image hashing scheme instead of cryptographic hashing techniques. The original image is divided into large blocks (sub-images) that are also divided into 8 x 8 blocks. Secure image hashing is utilized to generate the sub-image hash (signature) that may slightly change when the content-preserving manipulations are applied. Then, two sub-image hash copies are embedded into relatively-distant subbimages. The hash bits are robustly embedded in chosen DCT coefficients exploiting a property of DCT coefficients that is invariant with respect to JPEG compression. Furthermore, this technique is modified to be capable of compensating for the altered sub-images.
For authentication of video, the proposed fragile image authentication is modified to be appropriate for the requirements of video. In this video authentication techhnique, the approximated blocks and the blocks’ signatures of a frame are not inserted in the same frame but they are embedded into other frames. A modified version of this technique is also presented that reduces the computational complexity.