Elsevier

Displays

Volume 46, January 2017, Pages 52-60
Displays

Hierarchical recovery for tampered images based on watermark self-embedding

https://doi.org/10.1016/j.displa.2017.01.001Get rights and content

Highlights

  • A novel self-embedding watermarking scheme for tampering recovery is proposed.

  • MSB-layer bits are interleaved with distinct extension ratios to form reference bits.

  • Higher MSB layers have greater probabilities to be recovered than lower MSB layers.

  • Our scheme has better recovered results due to hierarchical recovery mechanism.

Abstract

In this paper, we propose a new self-embedding watermarking scheme with hierarchical recovery capability. The binary bits in the adopted MSB layers are scrambled and individually interleaved with different extension ratios according to their importance to image visual quality. The interleaved data, which are regarded as reference bits for tampering recovery, are segmented into a series of groups corresponding to the divided non-overlapping blocks, and then embedded into the LSB layers of blocks together with authentication bits of tampering detection. Because the extension ratios of MSB-layer bits are based on the hierarchical mechanism, the efficiency of reference bits is increased, and higher MSB layers of tampered regions have greater probabilities to be recovered than lower MSB layers, which can improve the visual quality recovered results, especially for larger tampering rates. Experimental results demonstrate the effectiveness and superiority of the proposed scheme compared with some of state-of-the-art schemes.

Introduction

In recent years, the rapid development of multimedia tools and Internet technology brings great convenience for transmitting and downloading multimedia data, which also leads to easier duplication and modification for digital contents than before in the meantime [1], [2]. Therefore, how to protect the security and integrity of the multimedia data [3], [4], especially the technique of image authentication [5], [6], [7], becomes an important research topic nowadays. The traditional image authentication scheme attaches digital signatures with the original image, and then compares the signatures of the received image with that of the original image to authenticate the integrity. However, it cannot locate and further recover the tampered region [8]. In order to solve these problems, fragile watermarking scheme for image authentication with the capability of tampering localization and content recovery has been proposed [9].

In the view of functions, fragile image watermarking schemes can be divided into two types. One type can just locate suspicious regions if the received image is tampered during transmission [10], [11], [12], [13], [14], [15], [16], [17]. This type of fragile watermarking schemes usually regards the hash of principal contents retrieved from each image block as its watermark data for embedding, and on the receiver side, the re-calculated hash of the received image is compared with the extracted hash of the image to detect the tampered regions, because tampering operation destroys the matching relationship between the contents of original image and the corresponding watermark data [10], [11]. In order to improve the accuracy of tampering detection, some researchers proposed the pixel-wise based fragile watermarking schemes. The watermark data derived from gray values of original pixels were embedded into the original pixels themselves, and then the tampered pixels can be located through the absence of watermark data [12], [13], [14]. In the scheme [15], a statistical mechanism was introduced into fragile image watermarking. The watermark data, including the tailor-made authentication data for each pixel and some additional test data, can be used to precisely locate the tampered pixels.

Another type of fragile watermarking schemes can not only locate the faked regions, but also can recover the located, tampered contents [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28], [29], [30], [31], [32], [33], [34]. In many practical applications, only tampering detection cannot satisfy the requirement, and the reconstruction for tampered regions is highly desirable. To achieve the self-recovery capability, Fridrich et al. conducted earlier attempt in this field, and they proposed fragile image watermarking scheme, which can realize content recovery after tampering detection [18]. They embedded the watermark of a block into the least significant bits (LSB) of other distant blocks, which was able to resist vector quantization (VQ) attack and collage attack. In order to accurately identify the faked image blocks, a digital watermarking method for image tampering detection and recovery was developed in [19], which was based on a 3-level hierarchical structure. This scheme can not only detect tampered areas accurately but also can deal with high tampering rate with acceptable recovered results. However, the above mentioned methods above cannot recover the tampered blocks whose watermarks embedded in other blocks were also destroyed, which was called as the tampering coincidence problem in Zhang et al.’s scheme [24]. Some watermarking schemes with self-correction capability were free of this problem. In [20], a fragile watermarking scheme with a hierarchical mechanism was presented, which can reconstruct original watermarked image without any error. The pixel-derived and block-derived watermark data were embedded into the LSBs of all pixels. This method had a limitation that the tampering rate must be no greater than 3.2% of the entire image to achieve the perfect restoration. In another work [21], an effective dual watermark scheme for image tampering detection and recovery was proposed by Lee and Lin. They applied two copies of watermark data for each block in the entire image, so it was able to provide the second chance for tampering recovery in case the first copy was damaged. However, the tampering coincidence problem still existed once both two copies of the embedded watermark data for the image block were destroyed. A self-embedding fragile watermarking scheme based on a reference sharing mechanism was proposed in [24], in which the watermark embedded into the three LSB layers of the whole image can be considered as the reference derived from the five most significant bits (MSB) layers of original image and shared by the whole image for further content restoration. As long as the content tampering was not too extensive, the five MSB layers of tampered regions can be perfectly recovered using the sufficient available data scattered in the intact blocks of image. Thus, it can effectively avoid the tampering coincidence problem. However, the way of reference data generation also caused the watermark wasting problem [26]. Huo et al. proposed an alterable-capacity fragile watermarking scheme in [28], which the watermark codes with the alterable-length consisted of three parts and were embedded into other three blocks. On the receiver side, two copies of significant-code were utilized to recover the tampered contents so that the recovery performance can be improved. However, this scheme was poor at dealing with the tampering form of random block missing.

In this work, in order to achieve better performance of visual quality for both watermarked image and recovered image, we propose a self-embedding watermarking scheme for tampering recovery based on hierarchical watermark embedding, which utilizes variable numbers of MSB layers to generate the shared reference data for content recovery and also variable extension ratios between the reference bits for each MSB-layer and the total reference bits for all adopted MSB layers. These parameters can be flexible according to different proportions of the tampered regions to achieve the satisfactory quality of recovered contents. During watermark embedding, the reference data are derived from each MSB layer, whose bits are interleaved and scrambled, and then are combined with the authentication data to form the watermark data to be embedded in the LSBs. Note that the proposed scheme is based upon the reference sharing mechanism and the extension ratio between the reference bits for each MSB-layer and total reference bits is variable. Thus, tampering coincidence problem is effectively avoided and the efficiency of watermark data can also be greatly improved.

The rest of this paper is organized as follows. Section 2 describes the procedure of watermark embedding, including watermark generation and data embedding. Section 3 presents the procedure of content recovery, including tampering detection and content recovery. Experimental results and comparison are given in Section 4. Section 5 concludes the paper.

Section snippets

Watermark embedding

The watermark embedding procedure of the proposed scheme consists of the following 3 stages: (1) Select the embedding parameters to generate reference data; (2) Generate authentication data using reference data with the embedding parameters; (3) Embed the watermark data, including reference data and authentication data, into original image to produce watermarked image.

Content recovery

After receiving the suspicious watermarked image, i.e., Iw, that may be damaged through the public channel, the receiver should first locate the tampered or missing blocks of Iw using the authentication bits, and then restore the MSBs of each detected, tampered block according to the reference bits and the MSBs retrieved from the intact blocks of the whole image.

Experimental results and comparison

All experiments were implemented on a computer with a 3.30 GHz Intel i3 processor, 4.00 GB memory, and Windows 7 operating system, and the programming environment was Matlab R2009b. A large number of test images sized 512 × 512 are used in our experiments to demonstrate the effectiveness of the proposed scheme. For color images, the luminance components were utilized for testing. In all following experiments, the subset length u of MSB bits for our scheme was set to 512, and the number m of MSB

Conclusions

In order to achieve better performance of tampering recovery, hierarchical reference-bits capacity according to the importance of image contents is utilized in the proposed scheme. The number of the MSB layers for the generation of the shared reference bits is variable, and the extension ratios between each MSB-layer bits and the total generated reference bits are different so that the efficiency of reference-bits can be increased. Through the hierarchical recovery mechanism, the higher MSB

Acknowledgments

This work was supported by the National Natural Science Foundation of China (61171126, 61272453, 61672354, 61303203, U1636101, 61401270), Ministry of Transport and Applied Basic Research Projects (2014329810060), the PAPD Fund, and the CICAEET Fund.

Fang Cao received the B.S. degree in applied electronics from Shanghai Normal University, Shanghai, China, in 2002, the M.S. degree in signal and information processing from Shanghai Maritime University, Shanghai, China, in 2004, and the Ph.D. degree in communication and information system from Shanghai University, Shanghai, China, in 2013. Since 2005, she has been with the faculty of the College of Information Engineering, Shanghai Maritime University, where she is currently a Lecturer. Her

References (35)

  • Z. Qian et al.

    Image self-embedding with high-quality restoration capability

    Digital Signal Process.

    (2011)
  • C. Qin et al.

    Self-embedding fragile watermarking based on reference-data interleaving and adaptive selection of embedding mode

    Inf. Sci.

    (2016)
  • Y. Huo et al.

    Alterable-capacity fragile watermarking scheme with restoration capability

    Opt. Commun.

    (2012)
  • C. Qin et al.

    Adaptive self-recovery for tampered images based on VQ indexing and inpainting

    Signal Process.

    (2013)
  • X. Qi et al.

    A singular-value-based semi-fragile watermarking scheme for image content authentication with tamper localization

    J. Visual Commun. Image Represent.

    (2015)
  • Z. Xia et al.

    A secure and dynamic multi-keyword ranked search scheme over encrypted cloud data

    IEEE Trans. Parallel Distributed Syst.

    (2016)
  • Z. Fu et al.

    Enabling personalized search over encrypted outsourced data with efficiency improvement

    IEEE Trans. Parallel Distributed Syst.

    (2016)
  • Cited by (0)

    Fang Cao received the B.S. degree in applied electronics from Shanghai Normal University, Shanghai, China, in 2002, the M.S. degree in signal and information processing from Shanghai Maritime University, Shanghai, China, in 2004, and the Ph.D. degree in communication and information system from Shanghai University, Shanghai, China, in 2013. Since 2005, she has been with the faculty of the College of Information Engineering, Shanghai Maritime University, where she is currently a Lecturer. Her research interests include image processing, computer vision and multimedia security.

    Bowen An received the M.S. degree in signal and information processing from Wuhan University, Hubei, China, in 2004, and the Ph.D. degree in Electronic Science and Technology from Chinese Academy of Sciences, in 2006. Since 2006, he has been with the faculty of the College of Information Engineering, Shanghai Maritime University, where he is currently a Professor. His research interests include remote sensing image processing and signal detection.

    Jinwei Wang was born in Inner Mongolia, China, in 1978. He received the Ph.D. degree in information security at Nanjing University of Science & Technology in 2007 and was a visiting scholar in Service Anticipation Multimedia Innovation (SAMI) Lab of France Telecom R&D Center (Beijing) in 2006. He worked as a senior engineer at the 28th research institute, CETC from 2007 to 2010. He worked as a visiting scholar at New Jersey Institute of Technology, NJ, USA from 2014 to 2015. Now he works as an associate professor at Nanjing University of Information Science and Technology. His research interests include multimedia copyright protection, image forensics, image encryption and data authentication. He has published more than 30 papers, hosted and participated in more than 10 research projects.

    Dengpan Ye was born in Hubei, China. He received the B.A.Sc in automatic control from SCUT in 1996 and Ph.D degree at NJUST in 2005 respectively. He worked as a Post-Doctoral Fellow in Information System School of Singapore Management University. Since 2012, he has been a professor in the school of computer science at Wuhan University. His research interests include Machine Learning and multimedia security. He is the author or co-author of more than 30 journal and conference papers.

    Huili Wang received the B.S. degree in communication engineering from University for Shanghai Science Technology, Shanghai, China, in 2014. She is currently pursuing the M.S. degree in signal and information processing from University of Shanghai for Science and Technology, China. Her research interests include data hiding, digital watermarking, and image authentication.

    This paper was recommended for publication by Pen-Cheng Wang.

    View full text