Passive video forensics has drawn much attention in recent years. However, research on detection of object-based forgery, especially for forged video encoded with advanced codec frameworks, is still a great challenge. In this paper, we propose a deep learning-based approach to detect object-based forgery in the advanced video. The presented deep learning approach utilizes a convolutional neural network (CNN) to automatically extract high-dimension features from the input image patches. Different from the traditional CNN models used in computer vision domain, we let video frames go through three preprocessing layers before being fed into our CNN model. They include a frame absolute difference layer to cut down temporal redundancy between video frames, a max pooling layer to reduce computational complexity of image convolution, and a high-pass filter layer to enhance the residual signal left by video forgery. In addition, an asymmetric data augmentation strategy has been established to get a similar number of positive and negative image patches before the training. The experiments have demonstrated that the proposed CNN-based model with the preprocessing layers has achieved excellent results.
K-pass pixel value ordering (PVO) is an effective reversible data hiding (RDH) technique. In k-pass PVO, the complexity measurement may lead to a weak estimation result because the unaltered pixels in a block are excluded to estimate block complexity. In addition, the prediction-error is computed without considering the location relationship of the second largest and largest pixels or the second smallest and smallest pixels. To this end, an improved RDH technique is proposed in this paper to enhance the embedding performance. The improvement mainly lies in the following two aspects. First, some pixels in a block, which are excluded from data hiding in some existing RDH methods, are exploited together with the neighborhood surrounding this block to increase the estimation accuracy of local complexity. Second, the remaining pixels in a block, i.e., three largest and three smallest pixels are involved in data embedding. Taking three largest pixels for example, when the difference between the largest and third largest pixels is relatively large (e.g., > 1), we improve k-pass PVO by considering the location relationship of the second largest and largest pixels. The advantage of doing this is that the difference valued 3 between the maximum and the second largest pixel which is shifted in k-pass PVO, is able to carry 1 bit data in our method. In other words, a larger amount of pixels are able to carry data bits in our scheme compared with k-pass PVO. Abundant experimental results reveal that the proposed method achieves preferable embedding performance compared with the previous work, especially when a larger payload is required. INDEX TERMS K-pass PVO, reversible data hiding, relative location, complexity measurement.
The main purpose of data hiding is to hide secret data imperceptibly into multimedia, ensuring secure data transmission. Existing magic matrix-based data hiding schemes cannot achieve satisfactory visual quality for a given low payload because they adopt a single-layer embedding (i.e., the fixed-sized reference matrix (RM) embedding), rather than adaptive multi-layer embedding (i.e., variable-sized RM embedding). To this end, we propose an adaptive data hiding scheme by constructing multi-layer RM using mini-Sudoku. Our scheme can adjust the number of layers (e.g., one, two, or more) of the RM to obtain the desired payload (i.e., low, moderate, or large payloads). In addition, our method also has the capability of obtaining the optimal number of layers of RM achieving the highest visual quality for a given payload. For each pixel pair, our scheme selects the optimal pair having the smallest Euclidean distance with this pair as its stego pair, resulting in a large decrease in distortion. Experimental results show that our proposed scheme can obtain various payloads, and more importantly, it can achieve better or comparable rate-distortion performance than several existing magic matrix-based data hiding schemes.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.