Search

Aiming at degrading the capability of the existing forensic methods in discriminating computer generated and natural facial images, a bidirectional conversion between computer-generated and natural facial images based on generative adversarial network (BDC-GAN) is proposed for anti-forensics in this paper. The generator of BDC-GAN is composed of noise encoding and content encoding. In the noise encoding, three high-pass filters are first utilized to extract the sensor pattern noise of the image, and then the stacked convolution layer is combined to continue encoding. In the content encoding, VGG-19 is truncated and fine-tuned to encode the content of the image. Some stacked convolution layers and adaptive instance normalization layer are used in the decoder. The discriminator uses multi-scale image discriminator. Furthermore, content loss and noise loss are well designed, and hyperparameters are reasonably set to accomplish the bidirectional conversion between two domain images meanwhile retaining the original facial contour. Experimental results and analysis demonstrate that the proposed anti-forensic method can achieve better visual quality and stronger deception ability compared with the existing unidirectional CG facial image anti-forensic methods and bidirectional domain adaptive methods, and its effectiveness is verified by the tests on the existing 9 forensic methods. It reveals that the existing forensic techniques can be bypassed by using adversarial learning, and it will eventually push the performance improvement of the discrimination of computer generated and natural facial images. [ABSTRACT FROM AUTHOR]

Aiming at resolving the problem of the irreversibility in some common neural networks for secret data extraction, a novel image steganography framework is proposed based on the generator of GAN (Generative Adversarial Networks) and gradient descent approximation. During data embedding, the secret data is first mapped into a stego noise vector by a specific mapping rule, and it is input into the generator of a GAN to produce a stego image. The data extraction is accomplished by iteratively updating the noise vector using the gradient descent with the generator. When the error is declined within the allowable error, the output image of the generator is approximate to the stego image, and the updated noise vector will also approach to the stego noise vector. Finally, the secret data is extracted from the updated noise vector. Experiments and analysis with WGAN-GP (Wasserstein GAN-Gradient Penalty) show that it can achieve good performance in extraction accuracy, capacity and robustness. Furthermore, the discussions also illustrate its good generalization with different GAN models and image datasets. [ABSTRACT FROM AUTHOR]

Aiming to reduce the distortion in the existing reversible watermarking for 3D mesh models, this paper generalizes two-dimensional (2D) region nesting to ${n}$ -dimensional space, and proposes a general region nesting based semi-fragile reversible watermarking for authenticating 3D mesh models by combing watermark generation based on vertex projection and mesh topology. For any ${n}$ -dimensional space, each hypercube space is partitioned into several nested watermark sub-spaces, and the watermark is embedded by placing the original vertex and the mapped vertex on a straight line. The watermark is extracted by acquiring the position of the vertices, and its original coordinates can be restored by inverse mapping. Furthermore, a new coordinate system is built for 3D mesh models based on reference vertices to achieve semi-fragility, and the watermarks generated by vertex projection and mesh topology is utilized for integrity authentication. Experimental results and analysis show that the proposed scheme can significantly reduce the distortion, achieve semi-fragility in RST (Rotation, Scaling and Translation) and vertex reordering, identify the tampering type, and accomplish tampering location at vertex level. It has great potential to be applied for content authentication of 3D mesh models. [ABSTRACT FROM AUTHOR]

To accomplish robust watermark extraction in reversible watermarking both in plaintext domain and encrypted domain, a separable robust reversible watermarking in encrypted 2D vector graphics is proposed in this paper. Firstly, a content owner uses a key to scramble the polar angles of the vertices to encrypt the graphics in the polar coordinate system. Consequently, a watermark embedder maps the encoded watermark segments to different vertices under the control of an embedding key and an HMAC (hash-based message authentication code) function. After that, the polar angle of the vertex is slightly adjusted to embed a watermark. Since the decryption operation does not affect the embedded watermark, the watermark can be extracted both in the plaintext and encrypted domain. Experimental results and analysis show that the proposed scheme can achieve good invisibility and reversibility. It can effectively resist normal operations such as rotation, scaling, translation (RST) and entity reordering, and it has good robustness against malicious attacks such as vertices/entities addition, deletion and modification. [ABSTRACT FROM AUTHOR]

Designing selective encryption (SE) schemes for H.265/HEVC has been attracted much attention with the advent of H.265/HEVC codec in the past two decades. However, the most SE algorithms for H.265/HEVC encrypt the syntax elements in the bypass mode to keep the bit rate. Moreover, the edge region of the video data is not sufficiently protected. To produce large visual distortion and edge loss, a tunable SE scheme for H.265/HEVC based on the chroma intra prediction mode (IPM) and coefficient scrambling is proposed. First, a pseudo-random number sequence is generated by AES-CTR. Then, the prediction, residual, and reconstruction information in the H.265/HEVC encoding process is encrypted by a pseudo-random sequence. It encrypts the syntax elements of context-based adaptive binary arithmetic coding (CABAC) in the bypass mode. Some syntax elements, including chroma IPM in the regular mode, are encrypted as well. To further protect the edge information, a coefficient scrambling is adopted. The edge information of each frame is extracted and the transform units (TUs) are classified according to it. Then, the coefficients of the TUs containing edge are scrambled. Finally, a sign used for marking the type of each TU is embedded into a coefficient. The experimental results and analysis show that the proposed scheme has better visual distortion and subjective evaluation results compared with some existing H.265/HEVC SE algorithms. Meanwhile, users can flexibly use the proposed SE scheme according to encryption performance and bit rate requirements, which is attractive in the scenario of protecting video in cloud servers. [ABSTRACT FROM AUTHOR]