Table of Content

30 September 2019, Volume 37 Issue 5

Previous Issue    Next Issue

Special Issue: Information Security of Multimedia

Survey on Analysis and Applications of Electric Network Frequency (ENF) Signals in Image and Video

CUI Sanshuai, MAO Maoyu, LIN Xiaodan, KANG Xiangui

2019, 37(5):  573-589.  doi:10.3969/j.issn.0255-8297.2019.05.001

Asbtract ( 932 )   PDF (7377KB) ( 284 )  

References | Related Articles | Metrics

Under the circumstance of alternate current (AC) electric field or around AC electrical equipment, multimedia recording equipment can capture the frequency fluctuation of electric networks simultaneously as it records video, image or audio programs. Electric network frequency (ENF) has the characteristics of real-time, continuity and consistency with surrounding electric networks, thus, the ENF information estimated from these multimedia files can be used for time stamping, information source detection, operation forensics, and media synchronization. In spite of different signal formats of ENF in audio, image and video programs, the detection and estimation methods of ENF are similar, and some processing methods are universal. Up to now, plenty of research works have been done on ENF in audio, but the works on ENF in image and video are far more behind. This paper presents a comprehensive review on the relevant researches, including characteristics, estimation and applications of ENF in images and videos. Firstly, the characteristics of ENF signal are explored by analyzing the processes of capturing images and videos. Then, the problems existing in the detection and estimation of ENF signal and corresponding solving approaches are summarized and analyzed. Finally, the applications of ENF in information forensics and the possible research directions in the future are discussed.

Evaluation and Comparison of Five Popular Fake Face Detection Networks

GAO Yifei, HU Yongjian, YU Zeqiong, LIN Yuyi, LIU Beibei

2019, 37(5):  590-608.  doi:10.3969/j.issn.0255-8297.2019.05.002

Asbtract ( 1220 )   PDF (13815KB) ( 552 )  

References | Related Articles | Metrics

Several fake face detectors based on convolutional neural network (CNN) have been reported to resist the impact of fake faces, but they all face a common problem that the intra-dataset test is generally with high accuracy, but the performance of crossdataset test drops significantly, which indicates low generalization ability. Based on thorough evaluations for five popular fake face detectors including MesoInception-4, MISLnet, ShallowNetV1, Inception-v3 and Xception, this paper completes both intra-dataset test and cross-dataset test on three fake face datasets. In experiment, the effects on generalization ability from of factors, such as dataset partition, data augmentation and threshold selection, are investigated.

Face Anti-spoofing Based on LBP, Multilayer DCT and CNN

LIU Wei, ZHANG Wanling, XIANG Shijun

2019, 37(5):  609-617.  doi:10.3969/j.issn.0255-8297.2019.05.003

Asbtract ( 1261 )   PDF (1734KB) ( 244 )  

References | Related Articles | Metrics

In applications of cyberspace security, face liveness detection is significant for face recognition systems. In this paper, an innovative face anti-spoofing algorithm against video-based face spoofing attacks is proposed based on local binary patterns and multilayer discrete cosine transform (LBP-MDCT) and convolutional neural network (CNN). First, we first extract face images from a target video, generate LBP features for each extracted face image and perform multilayer DCT onto the features to obtain LBP-MDCT features. Second, we input part of face images into CNN to obtain CNN features. After that, the two types of features are respectively input into support vector machine (SVM) classifier. In the last stage, the SVM output is fused with a decision-level operation to determine whether the target video is a spoof attack or a valid access. Compared with existing algorithms, the experimental results on two benchmarking datasets (Replay-Attack dataset and CASIAFASD dataset) demonstrate the excellent effectiveness of the proposed method.

Research on Facial Modification Detection Algorithm Based on Convolutional Neural Network

WANG Canjun, LIAO Xin, CHEN Jiaxin, QIN Zheng, LIU Xuchong

2019, 37(5):  618-630.  doi:10.3969/j.issn.0255-8297.2019.05.004

Asbtract ( 998 )   PDF (12455KB) ( 113 )  

References | Related Articles | Metrics

In order to avoid the influence of human factors on skin texture feature extraction of facial images, this paper proposes to detect facial image modification by using convolutional neural network (CNN) algorithm. To the best of our knowledge, this is the first report to use CNN in the detection of human face tampering. Compared with the traditional image classification methods which need complex artificial feature extraction, CNN can learn automatically, acquire features directly from the image, and reduce the difficulty of extracting features in traditional pattern recognition methods, accordingly, gaining a higher recognition rate and wider practicality at the same time. On the basis of the traditional convolutional neural network model, the proposed method builds a new network model for human face tampering detection by adjusting the size of the convolution kernel, reducing the parameters, changing the number of convolutional layer filters, adjusting the alternate mode of the convolutional layer and the pooling layer, and using dropout to improve the generalization ability of the model. Experimental results show that the new network model performs with a high recognition rate and strong robustness in the tamper detection of facial images.

Security Analysis of Image Encryption Algorithm Based on Bit Plane-Pixel Block Scrambling

QU Lingfeng, CHEN fan, HE Hongjie, YUAN yuan

2019, 37(5):  631-642.  doi:10.3969/j.issn.0255-8297.2019.05.005

Asbtract ( 877 )   PDF (11283KB) ( 166 )  

References | Related Articles | Metrics

In this paper, a known plaintext attack method based on the root mean square (RMS) of image block is proposed for ATBEM. Firstly, the scrambling sequence of bit planes is estimated to restore the original pixel value by using the invariant distribution ratio of 0 and 1 of plaintext images before and after encryption. Then, according to the characteristics of block scrambling and intra-block scrambling, which keeps the pixel value constant, a root mean square (RMS) feature of image block is defined to find and estimate the block scrambling matrix. The security performance of image block scrambling encryption is analyzed and discussed. The analysis results show that the smoother the image is, the smaller the blocks are, the more plaintext logarithms are needed to crack, and the more difficult the known plaintext attack is. Experimental results verify that the more texture the image is, the better the attack effect is. Under the condition of 2×2 block size, the attacker can crack more than 50% of the ciphertext image content with only one pair of known plaintext-ciphertext pairs known. And ATBEM encryption algorithm is difficult to resist the proposed known plaintext attack.

Hidden Voice Transmission Scheme Based on SSB Modulation

YU Xin, XU Zhengguang

2019, 37(5):  643-650.  doi:10.3969/j.issn.0255-8297.2019.05.006

Asbtract ( 851 )   PDF (6568KB) ( 89 )  

References | Related Articles | Metrics

In this paper, a hidden voice transmission scheme based on single sideband modulation (SSB) is proposed. Firstly, a band-stop filter is used to remove specific spectral components in background calls. Then, the voice that needs to be transmitted confidentially is embedded into the frequency domain cavity of normal voice in the form of SSB, thus realizing hidden voice transmission on smart phones. According to this scheme, the prototype system has been made to realize hidden voice transmission without changing the existing mobile phone software and hardware.

Image Splicing Localization Method Based on Fully Convolutional Residual Networks

WU Yunqing, WU Peng, CHEN Beijing, JU Xingwang, GAO Ye

2019, 37(5):  651-662.  doi:10.3969/j.issn.0255-8297.2019.05.007

Asbtract ( 1145 )   PDF (7143KB) ( 324 )  

References | Related Articles | Metrics

In order to solve the problem that the existing forgery localization networks are not easy to converge with the increase of the depth of networks, an image splicing localization algorithm based on fully convolution residual networks is proposed in this paper. On the one hand, the proposed algorithm transfers the idea of residual structure and introduces shortcut connection into part of convolution layers in fully convolutional network (FCN), so that the output is not a mapping input alone but the superposition of a mapping input and an input itself. On the other hand, conditional random field (CRF) is used as post-processing operation to improve splicing localization accuracy. Moreover, FCN and CRF are integrated in an end-to-end learning system. In addition, the proposed algorithm combines the prediction results of three kinds of FCNs (FCN8, FCN16 and FCN32). In our experiment, 5/6 of the spliced images in the public available dataset CASIA v2.0 are randomly selected as training set and the rest are used for testing. In order to test generalization performance of the proposed algorithm, the trained model is also cross-tested on another two public available datasets CASIA v1.0 and DVMM. The overall test results on three datasets show that the proposed algorithm performs better than some existing algorithms.

Steganalysis of Motion Vector-Based Steganography in H.264/AVC by Correlation Network Model

LI Songbin, YANG Jie, LIU Peng, WANG Lingrui

2019, 37(5):  663-672.  doi:10.3969/j.issn.0255-8297.2019.05.008

Asbtract ( 966 )   PDF (4445KB) ( 52 )  

References | Related Articles | Metrics

Motion vector modulated steganographic approach is an important type of information hiding method in H.264/AVC video streams, due to its large embedding capacity and low additional distortion induced in reconstructed video frames. In this paper, a novel video steganalysis algorithm is proposed for this type of information hiding method. Firstly, this paper designs a correlation network model to illustrate the correlation between temporal and spatial adjacent motion vectors. Secondly, we obtain a strong correlation network model by simplifying the original correlation network model through a pruning process, accordingly, the quantitative feature vectors of the strong model can be represented through quantifying the correlation of vertexes for steganalysis purpose. Finally, a steganographic detector based on the extracted feature vectors is built by using the support vector machine (SVM). Experiment results show that the proposed algorithm achieves a satisfying performance with the detection accuracy of more than 90%.

Improved Generation Camouflage Method Combined with Block Rotation and Photo Mosaic

ZHAO Yunying, SHAO Liping, WANG Yang, LU Hai

2019, 37(5):  673-690.  doi:10.3969/j.issn.0255-8297.2019.05.009

Asbtract ( 980 )   PDF (27886KB) ( 37 )  

References | Related Articles | Metrics

The conventional generation camouflage method combined with block rotation and photo mosaic expresses secret information by placing the circular images related to secret information in the embedding position, thus leading to a poor visual quality of resulted mosaic image and the leakage of hidden secret information. To address this, an improved generation camouflage method combined with block rotation and photo mosaic is proposed in this paper. In the method, firstly several grayscale images are converted into circular images and arranged as encoded images according to the ascending order of pixel mean value, and the cover image is directly transformed into the multiple halftone image by error diffusion. Secondly, the hidden positions of secret information are determined by random coordinate sequences, and the secret information was encrypted and mapped by generating random encryption mapping matrix. Finally, the coded images with the corresponding pixel values are selected by traversing the cover image. For the hidden and unhidden positions, the encoded images are rotated by a determined angle relative to secret information and a random angle respectively to generate a photo mosaic image. In order to improve the accuracy of secret information authentication, secret information is authenticated by the user key combined with centroid rotation matching strategy and interval authentication strategy. Experiments show that the proposed method directly uses the rotation angle of the encoded circular image to express secret information and it always selects the encoded circular image corresponding to the pixel value in the cover image to express secret information, so it neither brings any deviation nor causes any visual quality degradation. The proposed method performs strong anti-attack robustness and high security due to the entire key-dependency in extraction process.

Compact Image Hashing Algorithm Based on Opposite Color and Salient Region

ZHAO Yan, ZHOU Xiaowei, SHEN Qi

2019, 37(5):  691-703.  doi:10.3969/j.issn.0255-8297.2019.05.010

Asbtract ( 1003 )   PDF (17251KB) ( 84 )  

References | Related Articles | Metrics

In order to improve the recognition ability of algorithms by utilizing the color and local information of the image effectively, this paper proposes an image hashing algorithm based on color information and salient region. By reprocessing the input image, the proposed algorithm first obtains the color opponents' components and brightness components from the image. Then extracts the color features from the color opposition and the robust features of the salient areas from the image according to the visual attention weight matrix. Finally, the algorithm generates the final hashby combining and scrambling all these features. Experimental results show that the proposed algorithm performs with better image classification, shorter hash length and less computing time than the existing hash algorithms. Meanwhile, it also performs a good recognition ability in tampering detection application.

PRNU Extraction Algorithm Based on Trilateral Weighted Sparse Coding Model

ZHANG Yongsheng, TIAN Huawei, XIAO Yanhui, HAO Xinze, ZHANG Mingwang

2019, 37(5):  704-710.  doi:10.3969/j.issn.0255-8297.2019.05.011

Asbtract ( 824 )   PDF (3140KB) ( 96 )  

References | Related Articles | Metrics

Estimating the real noise of real-world image is the most important issue of image source forensics based on photo-response non-uniformity (PRNU). Compared with the estimation of additive white Gaussian noise (AWGN), most exsiting noise estimation algorithms used in PRNU extraction behave with poor satisfaction in estimating real noise. In this paper, we propose a PRNU extraction algorithm based on trilateral weighted sparse coding model (TWSCM). TWSCM has advantage in estimating the real noise of real-world image, because it can keep more PRNU noise in the estimation results. Having been tested on the largest image source forensics database, the proposed TWSCM-based PRNU extraction algorithm outperforms the existing algorithm of source forensic.

An Encrypted Traffic Identification Method Based on DPI and Load Randomness

SUN Zhongjun, ZHAI Jiangtao, DAI Yuewei

2019, 37(5):  711-720.  doi:10.3969/j.issn.0255-8297.2019.05.012

Asbtract ( 1129 )   PDF (10416KB) ( 96 )  

References | Related Articles | Metrics

With the development of encryption technologies and the emergence of private protocols, the identification of encrypted traffic has become an important research area in the field of information security. Based on the research of existing encrypted traffic identification technologies, an encrypted traffic identification algorithm based on DPI (deep packet inspection) and load randomness is proposed in this paper. The proposed algorithm mainly contains three steps. First, the DPI is used to filter and identify network traffic rapidly. Second, for those payload which could not be recognized by the DPI, their information entropies are calculated and the error of π-value is computed by Monte Carlo simulation. Finally, the C4.5 decision tree classifier is input for classification evaluation. The method can not only overcome the limitation that DPI can't fully identify the encrypted traffic and private protocol in the protocol interaction phase, but also solve the mis-distinguish of encrypted traffic and compressed file traffic as employing information entropy independently. Experimental results show that the proposed method is much more effective on encrypted traffic than the existing methods. At the same time, the method is proved to have good robustness.

Orthogonal GAN Information Hiding Model Based on Secret Information Driven

ZHU Yiming, CHEN Fan, HE Hongjie, CHEN Hongyou

2019, 37(5):  721-732.  doi:10.3969/j.issn.0255-8297.2019.05.013

Asbtract ( 1269 )   PDF (11129KB) ( 113 )  

References | Related Articles | Metrics

Driven by noise, the generator of the generative adversarial network (GAN) can generate high-quality digital images and provide a new data carrier for hiding information. In this paper, a coverless information hiding model combining orthogonal GAN and Ushape network is proposed on the fact that the orthogonal GAN discriminator can extract feature codes of the generated image. While hiding information, the binary sequence of the information to be hidden is mapped into a noise vector according to a group quantization rule, and the generator of the orthogonal GAN is driven by the noise vector to generate a hidden digital image. While extracting information, the feature code of the hidden image is extracted by the discriminator of the orthogonal GAN, and then the U-shape network is used to realize the mapping from the feature code to the driving noise, thereby recovering the secret information. By performing adversarial training of the built-in model with the CelebA dataset, the generator can generate high-quality hidden images and the discriminator can be combined with the U-shaped network to extract secret information from the hidden image. Compared with the latest similar algorithms, the proposed model performs better information extraction accuracy and security under the same information hiding capacity with reduced the training overhead and improved practicability.

Cover Selection Steganography Scheme Based on Image-to-Image Translation

LI Zonghan, LIU Jia, KE Yan, ZHANG Minqing, LUO Peng

2019, 37(5):  733-743.  doi:10.3969/j.issn.0255-8297.2019.05.014

Asbtract ( 840 )   PDF (12001KB) ( 94 )  

References | Related Articles | Metrics

To address the security risk in steganography algorithms using cover selection, an image-to-image translation steganography scheme based on cover selection is proposed. In the proposed scheme, the sender first selects the stego images from image libraries by the method of cover selection, then inputs the stego images into the cross-domain transformation model to get cross-domain images and transmits them to the receiver. Whereas the receiver receives the cross-domain images, and inputs them into the cross-domain transformation model to obtain stego images, then reversely converts them into secret information according to the mapping relation library. Experimental results show that the proposed method has significantly improvement in terms of capacity, anti-detection, and security.

Large Color Image Based High-Performance Reversible Data Hiding Hcheme for Various Capacities

WANG Junxiang, MAO Ningxiong, ZHAO Yi, WANG Chuntao

2019, 37(5):  744-760.  doi:10.3969/j.issn.0255-8297.2019.05.015

Asbtract ( 941 )   PDF (8336KB) ( 63 )  

References | Related Articles | Metrics

Reversible data hiding schemes generally show the advantage in integrity authentication, among which, histogram shifting (HS) technique is a special hot-spot due to its high effectiveness. However, HS processes generally employ empirical search criterions for side information (peak and zero bins) to reduce computation complexity, accordingly, resulting in the reduction of algorithm performance, such as low embedding capacity. To solve the problem, an adaptive side information selection method for dealing with various capacities (from low to large payloads) is proposed in this paper. It could determine the nearly optimal combination of side information based on the given payload. The proposed method employs a multi-feature based sorting technique is employed to search for smoother areas for data hiding, and designs an intelligent optimization algorithm to determine the optimal side information, thus resulting in a high performance. Experimental results show the proposed scheme could conduct information hiding for different quantities of payloads with superior performance.