Loading...

Table of Content

    31 March 2018, Volume 36 Issue 2
    Special Issue: Information Security of Multimedia Contents
    Reversible Data Hiding Based on Frequency Selection in JPEG Images
    HUANG Dan, CHENG Si-jin, HUANG Fang-jun
    2018, 36(2):  209-219.  doi:10.3969/j.issn.0255-8297.2018.02.001
    Asbtract ( 771 )   PDF (15216KB) ( 378 )  
    References | Related Articles | Metrics

    In this paper, by studying the encoding method of the discrete cosine transform (DCT) coefcients of joint photographic experts group (JPEG) image, and analyzing the relationship between the statistical properties of the DCT coefcients and coding length, we fnd that hiding data into DCT coefcients belonging to the mid-and-high frequency bands will lead to less fle size increment. Therefore, we propose a novel reversible data hiding (RDH) method based on the frequency selection strategy, in which the DCT coefcients of cover image are classifed into different groups based on their frequency bands, and the DCT coefcients of adjacent blocks belonging to the same frequency band are classifed into the same group. Thereafter, those groups with more zero coefcients are chosen preferentially for reversible data hiding. Experimental results demonstrate that the proposed RDH method can not only lessen the fle size increase introduced by data hiding greatly, but also perverse the image visual quality well.

    Reversible Information Hiding Method in Encrypted Image Based on Surface Interpolation
    CHEN Yan, YU Chun-qiang, HOU Xiao-jie, ZHANG Xian-quan, TANG Zhen-jun, HE Nan
    2018, 36(2):  220-236.  doi:10.3969/j.issn.0255-8297.2018.02.002
    Asbtract ( 919 )   PDF (4106KB) ( 362 )  
    References | Related Articles | Metrics

    This paper proposes a reversible information hiding method for encrypted images by means of surface interpolation. An original image is encrypted frstly, then a random function is used for obtaining the cover pixels in the encrypted image. Different hiding methods are adopted in accordance with the different inverted bits in the cover pixels. The stego-image is decrypted and the cover pixels are obtained. In the 5×5 neighborhood of a cover pixel, six non-cover pixels which are closet to the cover pixel are selected. Then surface interpolation is applied to calculate the predicted value of the cover pixel. The predicted value is used for extracting secret data and recovering the original pixel. The experimental results indicate that this method performs with a lower error rate in secret data extraction and a higher visual quality in image recovery.

    Improved Reversible Image Camouflage Method Based on Image Block Classifcation Threshold Optimization
    LIU Xiao-kai, YAO Heng, QIN Chuan
    2018, 36(2):  237-246.  doi:10.3969/j.issn.0255-8297.2018.02.003
    Asbtract ( 992 )   PDF (12684KB) ( 400 )  
    References | Related Articles | Metrics

    In order to improve the visual quality of stego images in digital image camouflage, a method of reversible image camouflage based on the threshold optimization of image sub-block classifcation is proposed. First, the sub-blocks of the original image and the cover image are classifed, respectively, according to their statistical characteristics. The threshold for classifcation is optimized through minimizing the mean square error of the camouflage image and cover image. Then, after the processes of the image sub-block matching, image sub-block linear transformation, sub-block rotation and horizontal flipping, a stego image which is visually similar to the cover image is generated. The transformation parameter information used for restoring the original image is eventually embedded into the stego image in a reversible manner to generate the fnal camouflage image. Therefore, the receiver side can extract the auxiliary information to realize the lossless recovery of the original image. The experimental results show that the visual quality of the camouflage image generated by the proposed method is better than that of the image without classifcation threshold optimization.

    Statistical Feature Hashing Based on Wavelet Decomposition
    SHEN Qi, ZHAO Yan
    2018, 36(2):  247-254.  doi:10.3969/j.issn.0255-8297.2018.02.004
    Asbtract ( 973 )   PDF (4605KB) ( 319 )  
    References | Related Articles | Metrics

    A statistical feature hash based on wavelet decomposition is proposed for the improvement of the image copy detection efciency and the recognition performance. In the proposal, an approximate image is frstly extracted from a preprocessed image by the threeorder wavelet decomposition. Secondly, the statistical features of the row and column of the approximate image of the third wavelet decomposition are extracted, and the L2 distance of the row and column statistical features is used as the invariant feature. All the invariant features are combined and used as the fnal hash of the image. Experimental results show that the proposed hash algorithm has better performance and higher efciency in copy detection.

    Synchronized Data Embedding and Scrambling Scheme in JPEG Images
    DAI Yu, YIN Zhao-xia
    2018, 36(2):  255-266.  doi:10.3969/j.issn.0255-8297.2018.02.005
    Asbtract ( 784 )   PDF (6733KB) ( 333 )  
    References | Related Articles | Metrics

    A synchronized data embedding-scrambling scheme in JPEG images is proposed in this paper. First, Checkerboard prediction method is used to predict the DC coefcient. According to the prediction error, the DC coefcient values are classifed into three categories:not-predicted (NP), predicted but not embedded (PN), and predictedand-embedded (PE). Then, the PE is vacated to embed information while scrambling the DC coefcients. Finally, the eligible RLC (run length coding) pairs of AC coefcients are rotated to embed information. Experimental results confrmed that the method ensures the high payload and the low fle expansion. The SSIM value of the reconstructed image is more than 0.996. The PSNR value of the reconstructed image is more than 50. And the original image can be reconstructed completely under certain conditions.

    Analysis and Improvement of a Reversible Hiding Algorithm in Encrypted Domain
    DENG Min, CHEN Fan, HE Hong-jie, YAN Shu
    2018, 36(2):  267-277.  doi:10.3969/j.issn.0255-8297.2018.02.006
    Asbtract ( 1002 )   PDF (5386KB) ( 562 )  
    References | Related Articles | Metrics

    It is analyzed that the four-adjacent-pixel prediction reversible information hiding algorithm cannot achieve a lossless recovery from encrypted images. In order to achieve the errorless recovery of the original image, a new the pixel-type is proposed by combining four neighborhoods with pixel to be predicted itself, and compressing type-mark matrix, which is used to record the smoothness of the pixels, and embedding it into the smooth pixels. Accordingly, in the recovery phase, the original image can be recovered perfectly with the help of the extracted type-mark matrix and its neighbor pixels. Experimental results demonstrate that the proposed method can achieve real reversibility, and further improve the embedding capacity and the quality of decryption image.

    P Frame PU Partitioning Mode Based H.264 to HEVC Video Transcoding Detection
    YU Li-fang, ZHANG Zhen-zhen, YANG Xian, LI Zhao-hong
    2018, 36(2):  278-286.  doi:10.3969/j.issn.0255-8297.2018.02.007
    Asbtract ( 1042 )   PDF (7683KB) ( 295 )  
    References | Related Articles | Metrics

    In this paper, a new algortithm engaging in detecting transcoding from H.264 to HEVC is proposed. PU partitioning mode, which is one of the new characteristics of HEVC (high efciency video coding), is investigated. The histogram of the PU size in the frst P frames of all GOPs is utilized as the feature set, and SVM with the statistical results is used for video classifcation. Experimental results demonstrate the effectiveness of our proposed method in distinguishing transcoded videos from single compressed videos, with a improved classifation accuracy of higher than 90%.

    Blind Detection of Audio Forgery Based on ENF Neighborhood Correlation Coefcient
    LÜ Zhi-sheng, TAN Li, FENG Bin, HU Yong-jian
    2018, 36(2):  287-298.  doi:10.3969/j.issn.0255-8297.2018.02.008
    Asbtract ( 1048 )   PDF (10633KB) ( 513 )  
    References | Related Articles | Metrics

    In order to improve the accuracy of the existing blind tamper detection methods based on electric network frequency (ENF) in the case of low SNR, we propose a novel blind detection approach based on the ENF cross-correlation coefcients in neighborhood. First, the ENF signal extracted from the query audio is divided into blocks, and the cross-correlation coefcients of the adjacent blocks are calculated. Then the adaptive fast transversal flter (FTF) is performed to the coefcient sequence. According to the variation of the fltering error energy, we can detect audio forgery. In order to reduce the interference and improve the accuracy of forgery detection and localization, the audio is processed in both forward and backward directions. Then the two directions'error energies are combined to detect forgery. Compared with two existing representative methods, the proposed method performs excellent accuracy both in forgery location and forgery detection. Especially under the circumstances of larger ENF fluctuation and lower SNR, the method shows more advantages.

    A Secure Image Retrieval Method Based on Combined Orthogonal Decomposition and BoVW
    ZHAO Xiao, XU Yan-yan, GONG Jia-ying, SONG Fang-zhen
    2018, 36(2):  299-308.  doi:10.3969/j.issn.0255-8297.2018.02.009
    Asbtract ( 801 )   PDF (5953KB) ( 477 )  
    References | Related Articles | Metrics

    We propose an image security retrieval method combining orthogonal decomposition and bag of visual word model (BoVW) in this paper to solve the problems that existing encryption algorithms for the current encryption domain image retrieval cannot meet the needs of different applications, and the algorithms relying on underlying image generally suffer lower retrieval accuracy. By introducing orthogonal decomposition framework, the image data domain is divided into the encryption domain and the retrieval domain. The encryption operation and feature extraction operation are independent, without interaction between them. In the encryption domain, users can choose any encryption method as needed. In the retrieval domain, the visual word bag model framework is introduced, and the image is represented as the visual word histogram, which reduces the semantic gap between the underlying feature and the high-level semantics, accordingly, improving the retrieval precision. Experimental results show that the proposed method provides higher security and higher retrieval precision than the current encryption domain image retrieval techniques.

    Preprocessing Layer in Spatial Steganalysis Based on Deep Learning
    SHI Xiao-yu, LI Bin, TAN Shun-quan
    2018, 36(2):  309-320.  doi:10.3969/j.issn.0255-8297.2018.02.010
    Asbtract ( 1084 )   PDF (5200KB) ( 687 )  
    References | Related Articles | Metrics

    In this paper, we propose some preprocessing methods to improve the performance of a well-designed convolution neural network based on the preprocessed layer. In the proposed methods, linear and nonlinear residuals are obtained by employing a set of derivative flters, and then quantized and truncated for the effective extraction. Experimental results show that the detection performances with the three proposed preprocessing methods are all improved. Although the improvements are not consistence under different spatial steganographic algorithms and different embedding rates. The detection performance is 6% better than the prior work for S-UNIWARD at 0.4bpp.

    Local Blur Detection of Digital Images Based on Deep Learning
    YANG Bin, ZHANG Tao, CHEN Xian-yi
    2018, 36(2):  321-330.  doi:10.3969/j.issn.0255-8297.2018.02.011
    Asbtract ( 1893 )   PDF (9689KB) ( 718 )  
    References | Related Articles | Metrics

    Blurring is generally a post-operation to conceal or remove the trace of tampering. In this paper, a new convolutional neural network model is proposed, and the corresponding network topology is presented to handle the problems in the detection of blur operations, such as Gaussian blur, average blur, or median blur. An information process layer is added into the conventional convolutional neural network to extract the residual features of fltering frequency domain, accordingly, improving the accuracy of blur detection between the frst-order and the second-order fltering operations. Experimental results demonstrate that the proposed method performs a higher accuracy in blur detection than traditional methods, and is able to discriminate between the common linear and nonlinear blur operations.

    Coverless Test Paper Disguise Combined with Non-direct Transmission and Random Codebook
    LU Hai, SHAO Li-ping
    2018, 36(2):  331-346.  doi:10.3969/j.issn.0255-8297.2018.02.012
    Asbtract ( 989 )   PDF (6123KB) ( 278 )  
    References | Related Articles | Metrics

    In traditional no-embedding, search-based coverless information hiding methods, there exist some shortages such as small capacity, requiring or maintaining a large database about texts or images with a high search cost. To address these problems, a coverless test paper disguise method combined with non-direct transmission and random codebook is proposed. Firstly, to avoid direct transmission of secret information, secret information is coded by random codebook which has been scrambled by pseudo-random sequence. Secondly, the coded secret information is converted into 32-decimal digit sequence and further expressed by 24-decimal and 9-decimal digit sequences. Finally, each 24-decimal and 9-decimal digits in these two sequences are disguised by the key based random offsets relative to option arrangement order in choice question and answer of blank-flling question respectively in a random generated test paper. Comparing with the existed methods, the proposed method avoids the direct transmission of secret information and does not require extra carrier modifcation by introducing direct stego test paper generation. The proposed method does not require a large database, not require any search cost. It depends on the correct key to obtain transmitted secret information with a high hiding capacity.

    Expansion of Video Forgery Detection Database and Validation of Its Effectiveness
    LI Ji-cheng, HU Yong-jian, Al-Alas Mohammed, XIONG Yi-chun, WEN Dong-xia, REN Yuan-yuan, LIAO Guang-jun
    2018, 36(2):  347-361.  doi:10.3969/j.issn.0255-8297.2018.02.013
    Asbtract ( 1281 )   PDF (21970KB) ( 264 )  
    References | Related Articles | Metrics

    Video forgery detection database VFDD1.0 effectively alleviates the current situation of lacking standard video forgery detection database, but still with drawback of insufcient capacity. To solve this problem, we expanded VFDD1.0 to VFDD2.0, which contains 1 550 videos, including 990 original videos captured with different imaging equipment and the corresponding 560 forged videos. In this paper, we describe the newly added videos and test the effectiveness of the expanded database with seven video forgery detection algorithms. Experimental results show that the proposed VFDD2.0 is capable of exhibiting performance of different algorithms more comprehensively, and is proved to be an effective database for video forgery detection.

    Adaptive Image Reversible Data Hiding Error Prediction Algorithm Based on Multiple Linear Regression
    WANG Xiao-yu, MA Bin, LI Jian, SHI Yun-qing
    2018, 36(2):  362-370.  doi:10.3969/j.issn.0255-8297.2018.02.014
    Asbtract ( 966 )   PDF (6567KB) ( 332 )  
    References | Related Articles | Metrics

    An adaptive error prediction method based on multiple linear regression algorithm to improve the reversible information hiding capacity is proposed. The inner relationship among pixels around the object pixel is learned adaptively based on the consistency feature of pixels distributing in local area of natural images, and a multiple linear regression function matrix is built to express the relationship. The object pixel is predicted accurately with the linear function learned from its neighboring pixels, rather than simply with the arithmetic combination of surrounding pixels. Experimental results show that the multiple linear regression based adaptive image error prediction algorithm can effectively enhance the reversible data embedding capability compared to other advanced error prediction methods.

    Coverless Information Hiding Based on Generative Adversarial Networks
    LIU Ming-ming, ZHANG Min-qing, LIU Jia, GAO Pei-xian, ZHANG Ying-nan
    2018, 36(2):  371-382.  doi:10.3969/j.issn.0255-8297.2018.02.015
    Asbtract ( 1724 )   PDF (3720KB) ( 1481 )  
    References | Related Articles | Metrics

    Traditional image steganography algorithms, which embed the secret information by modifying the content of the image more or less, are hard to resist the detection of image steganalysis tools. To address this problem, a novel coverless information hiding method based on generative adversarial networks is proposed in this paper. The main idea of the method is that the class label of generative adversarial networks is replaced with the secret information as a driver to generate hidden image directly. And the secret information is extracted from the hidden image through a discriminator. Experimental results show that this hidden algorithm ensures good performs in terms of steganography capacity, anti-steganalysis and safety.

    Adaptive Network Flow Watermarking Detection Scheme Based on Joint Centroid Entropy
    SHI Jin, LI Qian-kun, LIU Wei-wei, LIU Guang-jie, DAI Yue-wei
    2018, 36(2):  383-392.  doi:10.3969/j.issn.0255-8297.2018.02.016
    Asbtract ( 932 )   PDF (3938KB) ( 327 )  
    References | Related Articles | Metrics

    Considering the differences of watermarking in various types of complex network trafc, a new pre-grouping mechanism based on total packets number, average packets interval and bytes symmetry is designed. On this basis, an adaptive network flow watermarking detection scheme based on joint centroid entropy is proposed with the exploitation of the statistic variation of network trafc which is caused by interval-based flow watermarking. Experimental results on different types of trafc in anonymous communication system Tor show that the proposed method can achieve higher detection accuracy for random multi-key interval centroid based watermarking.

    Robust Watermarking of RGB Image Against Geometric Attacks
    LI Jing-xuan, XIANG Shi-jun
    2018, 36(2):  393-410.  doi:10.3969/j.issn.0255-8297.2018.02.017
    Asbtract ( 1179 )   PDF (43762KB) ( 269 )  
    References | Related Articles | Metrics

    In this paper, a robust watermarking algorithm based on RGB image is proposed for anti-geometric attacks by optimizing R, G and B channels. Firstly, the pixels in G and B channels of an image are expanded twice and three times, respectively. After that, the image is fltered by a Gaussian low-pass flter so that its mean and histogram can be used for watermark embedding. Comparing with the watermarking algorithm without pixel expanding in G and B channels, the proposed pixel-expanded algorithm shows stronger robustness. Furthermore, we embed the same number of bits into R,G and B channels for robustness testing, and the experimental results with 100 RGB images show that the strategy to expand the pixels in G and B channels is benefcial to improve the robustness of histogram-based watermarking for anti-geometric attacks.