1-4hit |
Wei ZHAO Pengpeng YANG Rongrong NI Yao ZHAO Haorui WU
Recently, image forensics community has paid attention to the research on the design of effective algorithms based on deep learning technique. And facts proved that combining the domain knowledge of image forensics and deep learning would achieve more robust and better performance than the traditional schemes. Instead of improving algorithm performance, in this paper, the safety of deep learning based methods in the field of image forensics is taken into account. To the best of our knowledge, this is the first work focusing on this topic. Specifically, we experimentally find that the method using deep learning would fail when adding the slight noise into the images (adversarial images). Furthermore, two kinds of strategies are proposed to enforce security of deep learning-based methods. Firstly, a penalty term to the loss function is added, which is the 2-norm of the gradient of the loss with respect to the input images, and then an novel training method is adopt to train the model by fusing the normal and adversarial images. Experimental results show that the proposed algorithm can achieve good performance even in the case of adversarial images and provide a security consideration for deep learning-based image forensics.
Nan LIU Yao ZHAO Zhenfeng ZHU Rongrong NI
This paper presents a commercial shot classification scheme combining well-designed visual and textual features to automatically detect TV commercials. To identify the inherent difference between commercials and general programs, a special mid-level textual descriptor is proposed, aiming to capture the spatio-temporal properties of the video texts typical of commercials. In addition, we introduce an ensemble-learning based combination method, named Co-AdaBoost, to interactively exploit the intrinsic relations between the visual and textual features employed.
Huawei TIAN Yao ZHAO Zheng WANG Rongrong NI Lunming QIN
With the rapid development of multi-view video coding (MVC) and light field rendering (LFR), Free-View Television (FTV) has emerged as new entrainment equipment, which can bring more immersive and realistic feelings for TV viewers. In FTV broadcasting system, the TV-viewer can freely watch a realistic arbitrary view of a scene generated from a number of original views. In such a scenario, the ownership of the multi-view video should be verified not only on the original views, but also on any virtual view. However, capacities of existing watermarking schemes as copyright protection methods for LFR-based FTV are only one bit, i.e., presence or absence of the watermark, which seriously impacts its usage in practical scenarios. In this paper, we propose a robust multi-bit watermarking scheme for LFR-based free-view video. The direct-sequence code division multiple access (DS-CDMA) watermark is constructed according to the multi-bit message and embedded into DCT domain of each view frame. The message can be extracted bit-by-bit from a virtual frame generated at an arbitrary view-point with a correlation detector. Furthermore, we mathematically prove that the watermark can be detected from any virtual view. Experimental results also show that the watermark in FTV can be successfully detected from a virtual view. Moreover, the proposed watermark method is robust against common signal processing attacks, such as Gaussian filtering, salt & peppers noising, JPEG compression, and center cropping.
A neighbor-aided authentication watermarking based on a chaotic system with feedback is proposed in this paper. This algorithm can not only detect malicious manipulations but reveal block substitutions when the VQ attack occurs. An image is partitioned into non-overlapped blocks. The pixels in one block and its neighboring block are combined to produce an authentication watermark based on a chaotic system with feedback, which is sensitive to the initial value. The produced watermark is embedded into the current block. During detection, the detector extracts the watermark firstly, then generates a reference sequence and compares it with the extracted watermark to authenticate the integrity of the image and locate the tampered regions. Experimental results prove the effectiveness of our method.