Dual watermarking framework for privacy protection and content authentication of multimedia

Highlights

• Dual watermarking based system for privacy protection and content authentication of multimedia.

• High robustness, offering threat identification and localization.

• Preservation of stego image’s quality via strong coefficient correlation between two selected neighborhood blocks.

• Highly robust to both singular and hybrid attacks.

Abstract

In the current multimedia networked infrastructure, privacy breaches due to cyber-attacks result in huge economic losses. Despite these threats there is ever increasing demand to share data over various insecure networks for accomplishment of numerous tasks. In such a scenario there is a greater need to develop new algorithms for strengthening the existing cybersecurity frameworks, ensure security, privacy, copyright protection and authentication of data. In this paper a new technique for copyright protection, data security and content authentication of multimedia images is presented. The copyright protection of the media is taken care of by embedding a robust watermark using an efficient inter-block coefficient differencing algorithm and is proposed as Scheme I. Scheme II has been utilized to ensure both copyright protection, and content authentication. The authentication of the content has been ensured by embedding a fragile watermark in spatial domain while as copyright protection has been taken care of utilizing a robust watermark. In order to thwart an adversary and ensure that it has no access to actual embedded data, we make use of a novel encryption algorithm in conjunction with Arnold transform to encrypt data prior to its embedding. The experimental results reveal that the proposed framework offers high degree of robustness against single/dual/triple attacks; with Normalized Correlation (NCC), more than or equal to 0.95. Besides, the fragile watermark embedding makes the system capable of tamper detection and localization with average BER more than 45% for all signal processing/geometric attacks. The average PSNR achieved for both schemes is greater than 41 dB. A comparison of the proposed framework with various state-of-the-art techniques demonstrate its effectiveness and superiority.

Introduction

Technology has made its way in almost every sector of life like commerce, defense, banking, movie industry, health-sector etc. Use of communication technology in the mentioned sectors involves data collection/transfer 24 $\times$ 7, from millions of individuals and devices. In several complex state-of-art systems huge data is shared in real time between different sensors and modules to carry out several automatic or monitored tasks [1], [2]. Such data may include health information, banking records and secret documents in the form of images, text files, audio and videos, acquired through various means and processed/stored by the different systems. The information exchanged is usually sensitive, as it comprises of private/secret data of individuals and organizations [3]. A minute alteration could result in misleading information affecting decision making of whole system. The presence of state-of-the-art multimedia editing tools are making it very easy to copy/edit and reproduce various multimedia objects. Given to this fact, a large number of data breaches caused due to various cyber-attacks have been reported in recent years. In US alone the number of data breaches arouse from 1093 million in 2016 to 1579 million in 2017. The total electronic records exposed in 2017 was found to be five time greater than 2016 [4]. These breaches proved quite costly in many cases as the companies lost a huge investment [5]. Electronic-healthcare sector of late has been a target of various data breaches. In July 2017 a cyberattack resulted in theft of medical images and details of patient visits, related to more than 12,800 patients of Arkansas Oral Facial Surgery Centre [6]. The Detroit health system reported a patient data breach in October 2017 due a cybercriminal theft of some email credentials wherein data of 18,470 patients was stolen [7]. Similar other multiple data breaches have been reported where privacy of millions of individuals has been exposed and thereby resulting in huge loses to the concerned organizations due to legal actions enforced by law agencies [8], [9], [10]. The healthcare organizations spend about $200 for every theft and more than $2 million for every data breach incident. The overall economic loss suffered by the US healthcare industry alone is about $5.6 billion per year, for rectification and management of lost and stolen records [11]. In India the scenario is no different and according to a latest report from Motion Picture Distributors’ Association (India), the Indian film industry loses around 2.7 billion dollars and over 60,000 jobs every year because of piracy [12]. A latest study reveals that the revenue at box-office would increase by 15% in case piracy is completely eliminated [13]. The online platforms like dark net and overlay nets are the easy to access, strong and unmonitored bases, for these hackers needing internet access and few high configuration processors. Similarly, several companies/agencies analyze the behavior of individuals sharing their data through social network services (SNSs), and use number techniques with the help of researchers to channelize their strategies accordingly [14]. Besides movie and healthcare industries, various data breaches have been reported in other industries [15], [16], [17]. So there is need to ensure extreme data security in all these areas. The focus of the cybersecurity techniques has been mainly in the sectors like e-banking, e-commerce, defense, multimedia industry, social media and e-healthcare. Cyber security is generally achieved using various technologies like hashing, steganography, watermarking and digital signatures etc. [18]. However, of late digital watermarking, has been proven, to be quite effective method for copyright protection and content authentication along with security and privacy of users.

Digital watermarking deals with hiding of secret data watermark) in some digital content which can be extracted only with the accurate knowledge of embedding procedure and some secret keys. Watermarking can be used by systems for applications like e-healthcare, defense, e-commerce, smart cities etc. [19], [20], [21]. Watermarking schemes are classified in many a ways [22]. Robust and fragile watermarking techniques are required for copyright protection and content authentication respectively. Some of the recently reported fragile watermarking schemes capable of informing the receiver whether the information received is tempered or not, could be seen in [23], [24]. The fragile watermarking schemes, usually implemented in spatial domain, offer least complexity and high embedding capacity but are less resistant to common signal processing attacks like compression [25]. Robust watermarking schemes, usually implemented in transform domain, in contrast focus on hiding the secret data in a cover media so that it could be recovered even if cover media is attacked [26], [27], [28].

In this paper, a blind dual watermarking framework using DWT, DCT, Arnold transform and a novel encryption algorithm has been proposed for privacy protection and content authentication of multimedia images. The block-based (DCT) transform avoids the blocking effects usually suffered by most watermarking schemes. This technique offers multiple options for the user to embed watermark as per various needs. Prior to embedding, Arnold transform and a novel encryption technique are applied to the watermark, to improve the security of the proposed watermarking framework. The secret encrypted data bits of the watermark are embedded into the DC coefficients of the image blocks, to achieve better robustness against the most of the signal processing attacks. As the proposed framework uses block processing procedure, it is flexible to adopt any of the image compression standards.

The main contributions of the proposed work are summarized as follows:

• Strong robustness is achieved by embedding the watermark bits in a way which ensures optimal difference between the modified coefficients. The difference between two coefficients is made to lie in four different zones where two zones are used for watermark bit ‘1’ and the remaining two are used to hold bit ‘0’. In addition, a sufficient guard band is used to separate the zones after embedding the bit ‘0’ and ‘1’. Since ample guard band is kept between the zones, any manipulation of the coefficients due to attacks fail to shift the difference in the wrong zone. The performance against geometric attacks is enhanced due to the application of Arnold transform which in addition to the security ensures that the neighboring bits of the watermark are embedded in distant blocks. So the effect of geometric attacks which target the particular portion of the image is minimized.

• The quality of the images is preserved by using only two coefficients out of 64 coefficients in an 8 $\times$ 8 block for embedding. Also an efficient coefficient correlation between two selected neighborhood block coefficients, ensures least modification done on the blocks due to embedding. The advantage of using two separate pair of zones is that, the modification of coefficients required for embedding watermark bit ‘1’ and ‘0’ is not too high. Thus ensuring good quality watermarked image.

• To ensure the tamper identification and localization, a fragile watermark is embedded in the spatial domain. Employing proposed fragile watermarking algorithm ensures that any minute tampering carried out on watermarked media effects the hidden watermark drastically.

• The effectiveness of the technique is tested for hybrid attacks in addition to singular attacks. This is important because, in real time scenarios the attacks on the image are usually multiple. The proposed algorithm performs well for dual attacks and even to most of the simultaneous triple attacks.

The rest of the paper is organized as follows. Section 2 describes the related work. Watermark preparation has been discussed in Section 3. Section 4 describes the watermark embedding and extraction algorithms. Experimental results and discussion have been presented in Section 5. The paper concludes with Section 6.