Generated text detectors can effectively detect the machine-generated texts which aim to produce false information to destroy the credibility of the media platform. However, generated text detectors are vulnerable to adversarial example attacks which focus on char-level perturbations to produce many word errors. In this paper, we design a sentence granularity based black-box attack model Sentence-Keyword-Attack (SK-Attack), which can effectively generate semantics-preserved, fluent, and grammatical adversarial examples. SK-Attack adaptively truncates the input examples based on sentence granularity and searches for the essential sentences to apply a sequence of contextualized perturbations with strict constraints. SK-Attack also applies keyword protection to preserve the keywords from being perturbed. Experiments show that SK-Attack outperforms the baselines when attacking the RoBERTa detector with various challenging generated text datasets and also has strong transferability to other attack models.

Frequent itemset mining is a data mining technique widely used on massive datasets. In cloud computing, the dataset may be encrypted for privacy protection. Therefore, frequent itemset mining over encrypted data is a crucial application in secure cloud computing. In this paper, we propose an effective privacy-preserving framework where the cloud server can directly perform data mining on the encrypted database without interacting with other cloud servers. We first design three security primitives to implement subset determination, accumulation, and comparison in the encrypted domain for frequent itemset mining. Based on the proposed framework, we then propose two secure protocols that allow the cloud server to perform frequent itemset mining on encrypted cloud data with these security primitives. The first protocol leaks no information to the cloud and the second protocol has the advantage of more efficient mining performance. We then present two strategies with parallel algorithms and GPU computing to accelerate the running time. We also analyze the security of our protocols and the computational complexities. Experimental results show that our serial-based protocols achieve shorter running times and higher levels of privacy than previous solutions. Our multi-CPU (or GPU) based parallel protocol can further reduce the practical running time.

Signal Processing in the Encrypted Domain (SPED) has received considerable attention as it aims at privacy-preserving solutions for various applications. Discrete cosine transform (DCT) is a popular signal transform widely used in signal processing. It has many applications in speech processing, still picture coding, image and video transformation, compression coding, and so on. In this paper, we mainly study how to implement DCT in the encryption domain with high precision. We propose a new scheme to implement encrypted domain DCT. This scheme encodes a complex number as a unit root polynomial in the evaluation, and realizes the high precision representation of complex numbers. With this representation, this scheme can also realize the high precision representation of decimals. To improve the computational efficiency, we also propose a fast implementation of DCT in the encryption domain, which can significantly improve the DCT speed for large-scale matrices. We conducted experiments to verify the effectiveness and efficiency. When the matrix size is small, our original method has the advantages of both high accuracy and fast speed. When the matrix size is large, the fast implementation outperforms the original method.

Line detection is an important research topic in image processing and computer vision. Hough transform is a widely used technique to detect lines. In the scenario of cloud computing, performing the Hough transform and line detection needs to consider privacy protection issues. In this paper, we propose implementing Hough transform in the encrypted domain and its application to line detection on encrypted images. Using the proposed encrypted domain Hough transform, we detect the extreme points in the encrypted parameter space. After transforming the extreme point into the encrypted spatial domain, we obtain the detected lines on the encrypted image. We conducted experiments and demonstrated the viability and effectiveness of our secure Hough transform and line detection on encrypted images.

Most state-of-the-art halftone image steganographic techniques only consider the flipping distortion according to the human visual system, which are not always secure when they are attacked by steganalyzers. In this paper, we propose a halftone image steganographic scheme that aims to generate stego images with good visual quality and strong statistical security of anti-steganalysis. First, the concept of pixel density is proposed and a novel construction called pixel density histogram (PDH) is proposed to design a “embedding” scheme for halftone images. Then, we optimize density pair selection to select density blocks that can improve visual quality. Finally, the messages are embedded through pixel density transition, where a novel pixel flipping strategy is proposed, which can maintain the structural dependence by optimizing the pixel mesh Markov transition matrix (PMMTM). The experimental results demonstrate that the proposed steganography scheme can achieve strong statistical security of anti-steganalysis with good visual quality without degrading the embedding capacity.

Some state-of-the-art binary image steganographic methods aim to generate stego images with good visual quality, while others focus more on the statistical security of the anti-steganalysis. This paper proposes a binary steganographic scheme that improves both of them by selecting more appropriate flipped pixels. First, a fused distortion measurement is developed that combines the advantages of flipping distortion measurement (FDM) and two data-carrying pixel location methods, including the edge adaptive grid method (EAG) and the “Connectivity Preserving” criterion (CPc). The FDM measures the distortion score by statistical features and achieves high-statistical security, while the EAG and CPc select pixels by analyzing the local texture structures based on visual quality. Then, to eliminate the interference brought by adjacent flipped pixels, a flipping position optimization strategy is proposed to find better positions for flipping pixels to further improve the steganographic performance. Experimental results have demonstrated that the proposed steganographic scheme can achieve stronger statistical security with better visual quality without degrading the embedding capacity.

Video watermarking has received much attention over the past years as a promising solution to copy protection. Watermark robustness is still a key issue of research, especially when a watermark is embedded in the compressed video domain. In this article, a robust watermarking scheme for H.264 video is proposed. During video encoding, the watermark is embedded in the index of the reference frame, referred to as reference index, a bitstream syntax element newly proposed in the H.264 standard. Furthermore, the video content (current coded blocks) is modified based on an optimization model, aiming at improving watermark robustness without unacceptably degrading the video's visual quality or increasing the video's bit rate. Compared with the existing schemes, our method has the following three advantages: (1) The bit rate of the watermarked video is adjustable; (2) the robustness against common video operations can be achieved; (3) the watermark embedding and extraction are simple. Extensive experiments have verified the good performance of the proposed watermarking scheme.

This paper proposes a content-based authentication scheme for tampering detection and localization of binary images. The watermark is generated by the feature vector of the original binary image, and embedded back into the image by a structural method. The feature vector we used is Zernike moments. It can tell the degree of the tamper in the binary image. A distance between two feature vectors is defined. We authenticate image by the distance between the extracted watermark and the feature vector of the test image. Once the tampering is detected, we can locate the tampered areas by comparing different components of the distance. The experimental results show that the algorithm can locate the tamper effectively.

This paper presents a novel energy-based watermarking scheme (EBW). The aim of EBW is to make the watermark survive video compression and collusion. After carefully analyzing the effect of the video compression on different type regions of video, two properties of the video coding are explored. According to these two properties, the watermark is deliberately embedded into the video to improve the robustness against video compression. At the same time, an improved fingerprint identification method is contained in EBW to help resist collusion attacks. The experimental results demonstrate the validity of EBW.

Our previous paper [1] proposed to use Zernike moments magnitudes (ZMMs) of the image for authentication. We demonstrated the good robustness and discriminative capability of this feature vector and proposed a structure embedding method for the watermark generated from this feature vector to locate the tampered area. In this paper, we extend and improve the algorithm in [1] in the following ways: 1) embedding watermark in an image will affect its feature vector, we split the original image into several spaces randomly, one for generating feature vector, one for embedding ZMM based watermark. Thus, we can remove the effect of watermarking itself on the feature vector generating, 2) presenting a two-stage authentication method and improve the discriminating capability of the authentication. Compared with the existing approaches, the proposed scheme has better performance of discriminating high quality JPEG compression from malicious manipulations.

In this paper, we propose a content-based binary image authentication scheme. At first, we use Zernike moments magnitudes (ZMM) to generate the feature vector and demonstrate that this feature vector can represent the binary image and decide its authenticity effectively. Then the watermark is generated by quantizing ZMMs and embedded into the image. The authentication doesn't need the original watermark. The decision depends on the distance between the extracted watermark and the feature vector of the test image and a metric measure. To decrease the influence of watermarking on the feature vector, we split the binary image into two parts by a random mask, one for generating feature vector and the other for embedding watermark. Zernike moments are usually computationally expensive, so we propose a fast algorithm. Extensive experiments show that our scheme can detect malicious attacks effectively.

In this paper, we present a hybrid watermarking scheme for video authentication based on wavelet domain. It embeds one robust watermark for temporal authentication distinguishing different inter-attacks, such as frame loss, inserting and reordering. Other two watermarks are used for intra authentication, one for discriminating the malicious attacks from acceptable manipulations, and the other for locating malicious attacks. The latter two watermarks are content-based. Zernike moments magnitudes (ZMMs) of the lowpass wavelet band of the host video frames are chosen as features. Experimental results show that ZMMs are robust to MPEG-2 compression and slight noise, while fragile to content change. This semi-fragile property is used to tell malicious attacks from non-malicious attacks. We also use structure of the embedded ZMMs to locate the tampered area. Experimental results show that this scheme can authenticate the video effectively.

In this paper, a new embedding strategy for DWT-based watermarking is proposed. Different from the existing watermarking schemes in which low frequency coefficients are explicitly excluded from watermark embedding, we claim that watermarks should be embedded in the low frequency subband firstly, and the remains should be embedded in high frequency subbands according to the significance of subbands. We also claim that different embedding formula should be applied on the low frequency subband and high frequency subbands respectively. Applying this strategy, an adaptive algorithm incorporating the feature of visual masking of human vision system into watermarking is proposed. In the algorithm, a novel method to classify wavelet blocks is presented. The experimental results demonstrate that the watermarks generated with the proposed algorithm are invisible and robust against noise and commonly used image processing techniques.

Abstract: Robustness is one of the major issues of digital watermarking algorithm. An effective method to improve the robustness of the watermark is to embed the watermark adaptively based on the perceptual property and signal characteristics. In this paper, an adaptive video watermarking algorithm based on wavelet domain is proposed. According to the properties of the 2-D wavelet coefficients, the watermark is inserted into the low frequency subband coefficients to achieve better robustness. In order to improve the strength of the components of the watermark, we propose to classify the coefficients of the low frequency subband based on the motion of the object and the texture complexity of the content in the video sequence. According to the result of classification, the strength of the watermark component is adjusted adaptively. The experimental results show that the watermark just generated is robust to video degradation and distortions, e.g., those that result from additive Gaussian noise, MPEG-2 coding at high compress-ratio, temporal downsampling and spatial downsampling, while the transparency of watermark is guaranteed.