Elsevier

Signal Processing

Volume 98, May 2014, Pages 52-61
Signal Processing

Reversible joint fingerprinting and decryption based on side match vector quantization

https://doi.org/10.1016/j.sigpro.2013.11.011Get rights and content

Highlights

  • We propose a joint fingerprinting and decryption (JFD) based on side match vector quantization (SMVQ) for transmitting confidential information and traitor tracing in multimedia distribution.

  • The proposed scheme can extract the fingerprint without referring to the original image.

  • The fingerprinted copy can be recovered to the un-fingerprinted version after the extraction of the fingerprint bits.

  • The proposed scheme can achieve high perceptual security (with PSNR below 12 on average), high fingerprinted visual quality (with PSNR above 28 on average), and desirable fingerprint payload (about 0.6 bit/per block).

Abstract

This paper proposes a joint fingerprinting and decryption (JFD) based on side match vector quantization (SMVQ) for transmitting confidential information and traitor tracing in multimedia distribution. With the help of SMVQ, the proposed scheme can extract the fingerprint without referring to the original image. In addition, after the extraction of the fingerprint, the fingerprinted copy can be completely recovered to the un-fingerprinted version. The experimental results show that the proposed scheme can achieve high perceptual security (with PSNR below 12 on average), high fingerprinted visual quality (with PSNR above 28 on average), and desirable fingerprint payload (about 0.6 bit/per block).

Introduction

The protection of multimedia content against illegal distribution has become increasingly important due to advances in multimedia techniques and communication technology that provide users with rapid access to very large volumes of multimedia data. Consequently, various multimedia security methods have been proposed. For example, to protect confidentiality, multimedia encryption methods [1], [2], [3], [4] are generally used to make the original content unintelligible and only the authorized user can recover it successfully. Digital watermarking methods [5], [6], [7], [8] are another approach to protecting integrity and ownership. These watermarking methods securely embed an authentication code, or watermark, in the original data. The embedded watermark is invisible and can only be extracted by the rightful owner.

Although the aforementioned methods can achieve confidentiality, integrity, and ownership verification, protecting multimedia data from being illegally disseminated is often neglected. Copies of decrypted multimedia data being redistributed by authorized users to unauthorized users lead to pirates running wild. Finding such authorized users is referred to as a traitor tracing problem. To meet this challenge, better distribution schemes called fingerprint-based schemes [9], [10], [11], [12] have been developed and proposed, especially for tracing a traitor in secure multimedia communication. These schemes are based on digital fingerprinting which embeds a fingerprint in subscribed data. Thus, when different customers subscribe to the same multimedia data, they will receive different copies (due to different fingerprints), but with the same visual content.

In general, there are two types of fingerprinting approaches: server-side fingerprinting and receiver-side fingerprinting. The latter is more practical since only one media stream is required to be disseminated to users in a multicast manner [11]. On the contrary, if server-side fingerprinting is used [9], [10], the sender needs to consume a lot of bandwidth for sending different copies to different users. Therefore, this paper focuses solely on receiver-side fingerprinting.

Macq and Quisquater [12] proposed the first receiver-side fingerprinting for the protection of digital content in order to reduce computation and bandwidth cost on the server. At the receiver side, however, decryption and fingerprinting are performed independently, as shown in Fig. 1, and thus, the original content may be intercepted by attackers at the gap between the decryption and fingerprinting operations. An alternative digital content protection method of the receiver side embedding is known as a joint fingerprinting and decryption scheme (JFD). The JFD schemes [13], [14], [15] combine fingerprint embedding and decryption, as depicted in Fig. 2, to prevent an attacker from obtaining the plain image. The Chameleon method for traitor tracing, proposed by Anderson and Manifavas [13], encrypts the original data at the sender side, and decrypts it using a simple least significant bit (LSB) substitution technique. As a result, the decrypted data provides good perceptibility, but this method is not robust to signal processing such as recompression, additive noise, and filtering. Kundur and Karthik [15] proposed an alternative JFD scheme based on partial decryption, in which the signs of DCT coefficients are fully encrypted at the sender side but not all the encrypted signs will be decrypted. These unchanged signs determine the user's fingerprint. The drawback of the scheme is that the encrypted image is intelligible. Lemma et al. [14] also proposed a JFD method through partial encryption. Their scheme encrypts the media data at the sender side, and decrypts it with a different key stream. Each key stream has the same size as the media, leading to prohibitive transmission cost. A generalized version of Lemma et al.’s method has been proposed by Celik et al. [16], which not only reduces transmission costs but also provides an efficient fingerprint detection method and rigorous security proofs. Although the above JFD methods meet the requirements of protecting media distribution, they are performed on either the transform domain or the pixel domain, and none of them can be applied to the vector quantization (VQ) domain. Recently, Lin et al. [17] proposed the first JFD scheme based on vector quantization. This method extends the JFD function such that it is resistant to noise interference.

There is still a significant issue needs to be resolved for the previous JFD methods. That is, when a fingerprint is extracted from a fingerprinted image, the original image needs to be present. If the original image is lost or severely damaged, the content owner will have difficulty finding the traitors. Therefore, it is important that the original content can be recovered from the fingerprinted image without any error in order to verify the unique customer ID, or fingerprint. In addition, the reversibility can be used to verify whether the fingerprinted image has been tampered with. If the fingerprinted image has been changed, it cannot be recovered to the original image. As a result, a reversible JFD scheme is necessary for fingerprint extraction and tampering detection.

In this paper, we propose a reversible JFD scheme based on side match vector quantization (SMVQ). In the pre-processing step, the proposed scheme transforms a VQ image into a SMVQ image. Then, a codeword substitution technique is applied to scramble the image. At the receiver side, the steps of decryption and fingerprinting are done simultaneously and the fingerprinting process is performed by the SMVQ approach. The proposed scheme can extract the fingerprint from a fingerprinted image without the help of the original content, which makes the proposed scheme more efficient than previous JDF schemes. Furthermore, even if the original image is lost, the proposed method can recover the original SMVQ content after extracting the fingerprint, which greatly enhances the robustness of the proposed method.

The rest of this paper is organized as follows. In Section 2, VQ and SMVQ methods are presented. Section 3 details the proposed JFD scheme based on side match vector quantization (SMVQ). Experimental results and discussions are given in Section 4. Finally, conclusions are summarized in Section 5.

Section snippets

Related works

In this section, we shall first introduce the basic concept of VQ followed by the SMVQ technique that is mainly used in improving VQ compression rates. Our previous JFD work is also covered at the end of this section.

The proposed method

Our proposed JFD scheme, as shown in Fig. 5, utilizes a reversible embedding of SVMQ-based compression technique to enhance the performance of the JFD scheme in the fingerprint extraction process. In other words, the server can extract fingerprint codes from fingerprinted copies without the help of the original images. To achieve our goal, an original VQ image is modified by new indices from the SMVQ encoder. After the modification, the modified image P is then encrypted using codeword

Experimental results and Disscussions

In the experiments, six standard 512×512-pixel grayscale images are used and modified by our SMVQ approach, as shown in Fig. 7 (“Lena”, “F16”, “Baboon”, “Boat”, “Pepper”, and “Barb”). Each image is divided into 16,384 image blocks of sized 4×4 pixels. Three different sizes of codebooks, 256, 512, and 1024, are used to evaluate the proposed scheme. The size of the sub-codebook used in SMVQ is set to 16 in our experiments. Our codebooks in the experiments are trained by the LBG (Linde-Buzo-Gray)

Conclusions

In this paper, we propose an extension to JFD in the vector quantization domain. The method encrypts the content using a supper codebook before transmission. At the client side, decryption is performed using a sub-codebook generated by the SMVQ process. The decryption will also leave a unique fingerprint on the content. Experimental results show that the fingerprinted image quality is only slightly inferior to that of our previous JFD method [17] due to the use of SMVQ. However, the proposed

Acknowledgments

This work was supported by National Science Council, Taiwan, under Grants NSC 101-2221-E-468-021.

References (25)

  • F. Hartung, B. Girod, Digital watermarking of MPEG-2 coded video in the bitstream domain, in: Proceedings of IEEE...
  • S. Lian

    Multimedia Content Encryption: Techniques and Applications

    (2008)
  • Cited by (10)

    • Joint fingerprinting and decryption method for color images based on quaternion rotation with cipher quaternion chaining

      2016, Journal of Visual Communication and Image Representation
      Citation Excerpt :

      The unique feature of this method is blind detection, i.e. the fingerprints can be extracted without the original image, which is not common among fingerprinting methods. Recently, several JFD methods based on vector quantization (VQ) have been proposed [23–26]. The first method [23] encrypts images via permutation and codeword substitution by using static key-trees.

    • Secure multicasting of images via joint privacy-preserving fingerprinting, decryption, and authentication

      2016, Journal of Visual Communication and Image Representation
      Citation Excerpt :

      Nevertheless, the drawback of this method is if the original image is lost or severely damaged, the fingerprint cannot be extracted and the tampered area cannot be identified. This led to the extension work as discussed in [19]. Recently, Czaplewski and Rykaczewski [9] presented a JFD method that utilizes a simple block cipher based on matrix multiplication to encrypt original images.

    • Matrix-based robust joint fingerprinting and decryption method for multicast distribution of multimedia

      2015, Signal Processing
      Citation Excerpt :

      The unique feature of this method is blind detection, i.e. the fingerprints can be extracted without the original image which is not common among fingerprinting methods. Recently, a number of JFD methods based on vector quantization (VQ) have been proposed [19–22]. The main reason for choosing vector quantization was its very strong tolerance to noise interference.

    • Joint fingerprinting and encryption in hybrid domains for multimedia sharing in social networks

      2014, Journal of Visual Languages and Computing
      Citation Excerpt :

      However, the encrypted data is not secure in visual perception because the encryption of signs of DCT coefficients cannot fully scramble the original data. A JFD scheme based on vector quantization is proposed in [14–16]. In [17], the JFE scheme in the compressed domain is proposed.

    View all citing articles on Scopus
    View full text