1 Introduction

Digital data has invaded all kinds of public media such as video, image, audio, and text. Digital data are transmitted frequently over the Internet. Efficient global computer networks have acted as a catalyst for the ever-growing demand for digital media. Hence, digital data are more susceptible to attacks and can be easily compromised [4, 5, 47]. With the help of watermarking, it is possible to identify the (creator, owner, distributor or approved consumer) of an image or document [37, 48, 59]. It can also aid in detecting whether the image or document has been tampered with. Developing a robust watermarking technique having high computational efficiency is the need of the hour. There is always a trade-off between the features of watermarking. Existing methodologies are primarily concerned about the robustness, imperceptibility and embedding capacity. Security and complexity features are given less priority in the development of the watermarking scheme. Nowadays powerful multimedia editing software has invaded the market, thus increasing the gravity of malicious media content. Methods like cryptography and steganography can protect digital content. However, one of the most efficient countermeasures against malicious data is digital fragile watermarking [23]. However, most of the existing methods focus on embedding capacity, robustness and losing sight of security. For the security of watermark, the encryption technique can be used [52, 54]. During the last few years, the multimedia watermarking scheme has been evolved rapidly. Apart from this, watermarking is also evolved in the area of IP protection, relational database, cyber-physical system, IOT & 5g technology, and e-governance [9, 24, 25, 58, 72, 76, 79, 80, 83]. Watermarks should be sturdy against information manipulations (like advanced arrange transformation and reckoning digital-to-analog conversion). The following elementary conditions in watermarking apply to all or any media: i) A watermark ought to provide the maximum amount of knowledge as conceivable, which means the watermark data rate must be high, ii) A watermark must stay inside the host data no matter anything happens to the host data. This necessity is stated as robustness, iii) A watermark must be irremovable, iv) A watermark must be common, secret as well as accessible to the authorized party.

During, the last decades, a number of the comprehensive surveys have been published in the area of digital watermarking [2, 32,33,34, 36, 40, 87, 102, 109]. This survey focuses on the various features and application of the watermarking. In this work, we had done a comprehensive survey based on the multimedia type. Key features of this survey include-

  1. i)

    Comprehensive literature review of watermarking based on different multimedia types.

  2. ii)

    Comparative analysis in terms of techniques used and the purpose of the proposed watermarking scheme.

  3. iii)

    This literature review also deals with the different features and its trade-offs, additionally, it deals with the various attacks on watermarking which will helps the researcher to design an optimal watermarking scheme.

  4. iv)

    This paper will also help the researchers in a concerned field to analyses that which technique is best in terms of that particular application, attacks, and features.

The outline of the rest of the paper is given below. Section 2 presents the basic concept of watermarking and its classification. Features of watermarking are section 3. In Section 4 State-of-the-art of watermarking techniques is presented. Section 5 deals with the various performance measure of watermarking. Several attacks of watermarking are discussed in Section 6, and finally, we conclude our works in Section 7.

2 Digital Watermarking and its Classification

Digital Watermarking is a method used to set up the identity of any information to protect it from any kind of illegal alteration or use [13, 38, 85]. Digital watermarking process involves two steps - i) Watermark Embedding - in this step the watermark is inserted into the host signal by utilizing the defined algorithm and ii) Watermark Extraction - in this step the watermark is extracted from the watermarked signal.

The watermarking methods are often classified into varied categories. Several authors have classified the watermarking approach in various classes. Fig. 1 presents the various classification of the watermarking. In this work we had classified the watermarking in the three classes - i) classification based on multimedia, ii) classification based on characteristics and iii) classification based on application

Fig. 1
figure 1

Classification of Digital Watermarking

Full size image

2.1 Classification based on Multimedia

Digital text watermarking: - Text data consist of different semantic entities, such as word, sentence, row, paragraph and punctuation mark, etc. The role of syntax and semantics is quite important here, all trans-formation required here is related to one of them to embed the digital watermark into cover text (i.e. text in which watermarks have to be embedded).

Audio watermarking: - Audio watermarking is more challenging than image and video watermarking, since the Human Auditory System (HAS), is notably more sensitive than the Human Visual System (HVS).

Image watermarking: - Image is generally of large size compressed, highly robust & imperceptible watermarks are required to embed into the cover image.

Video watermarking: - Due to 3D characteristics of the video, imperceptibility of the watermark is difficult in video watermarking. The other issues with video watermarking are that video signals are extremely vulnerable to pirate attacks.

Graphic watermarking: - 2D or 3D computerized graphics can have a watermark embedded to indicate the copyright protection.

Database watermarking: - Database can gain more security in terms of confidentiality, integrity, etc. by embedding a digital watermark in the database. To deploy database watermarking, particularly in the healthcare domain, distortion caused due to the watermark should be least to ensure the correct interpretation of records.

2.2 Classification based on Characteristics

Based on characteristics the watermarking methods will be of the subsequent sorts i.e. i) Non-Blind & Blind, ii) Imperceptible & Perceptible, iii) Public & Private, iv) Robust, Fragile & Semi-Fragile digital watermarking, v) Frequency & Spatial Domain

  1. i)

    Blind and Non- Blind digital watermarking: -Based on the watermark extraction, watermarking technique can be classified into 3 classes like blind, semi-blind, and non-blind. In blind watermarking the watermark is extracted without the cover/original data.

  2. ii)

    Visible & Invisible Watermarking: - A watermark that is visible to the human eyes is called visible watermarking, otherwise the watermark is said to be invisible. Visible and invisible watermarking techniques are also known as perceptible and imperceptible watermarking respectively.

  3. iii)

    Private and public digital watermarking: - if a watermark is detectable only by authorized users than it is said to be a private digital watermark. When compared with a public digital watermark it is found that private digital watermarking is more robust. If a watermark is detectable by anyone then it is said to be a public digital watermark. There is another form of it known as asymmetric form, through that without disturbing watermark any user can read it. In this case, verification is done by public key & embedding done using the private key.

  4. iv)

    Robust, Semi-fragile and Fragile digital watermarking: - if watermarks survive malicious (i.e. that destroys watermark) and non-malicious (do not explicitly mean to modify it) attacks then those watermarks are known as a robust watermark. Copyright protection is one of the applications were using this kind of digital watermark will be beneficial as they are more prone to malicious & non-malicious attacks. A semi-fragile watermark is proposed to sense any unauthorized alteration, & parallel allowing some image processing operations. They find their application in some selected authentication techniques. Even a slightest change (intentional/unintentional)in watermarked image can be detected by the fragile watermarking scheme.

  5. v)

    Frequency & Spatial Domain-based: - The watermark is deep-rooted in the cover image (data) by neutering the gray-scale value of the pixels of the novel image (data) while not applying any conversion in case of the spatial domain. While in frequency domain digital watermarking can be done using some transformations such as DWT, Discrete Cosine Transformation (DCT), Discrete Fourier Transformation (DFT), etc. Spatial domain techniques are less robust than the Frequency domain.

  6. vi)

    Adaptive & Non-Adaptive: - In adaptive, the embedded data is varied on the basis of local properties of host data. Varying the embedded data includes locally adjusting the amount of embedding power and/or controlling the locations where embedding is to be done. Whereas, in the non-adaptive techniques global properties of the host data are used to control the global embedding parameter.

2.3 Classification Based on Application

  1. i)

    Copyright Protection: - One of the major drawbacks of digital multimedia is that it is prone to easy illegal copying techniques like piracy. So, the need for the techniques to protect the copyright of digital data is needed.

  2. ii)

    Data Authentication: - A digital signature is one of the outstanding cryptanalytic methodologies for data confirmation. In any case, just in case of the loss of the signature, the confirmation work couldn’t be performed. The solution for this is that the signature will specifically be inserted in the work utilizing watermarking. A viable authentication arranges must have the capability to manage whether or not an image or document has been adjusted or not, ready to distinguish any modification created on the image or text.

  3. iii)

    Fingerprint: - As a result of the progression of innovation, one of the conceivable uses of digital watermarking is fingerprinting. Fingerprinting in digital watermarking is for the most part utilized as the way toward inserting the uniqueness to the image with the end goal that it is hard to temper or abrogate. This allows the copyright holder to find a freebooter if the image is circulated illicitly.

  4. iv)

    Copy Control: - Copy control is limitation of the advanced media. It can be duplicated without degrading the original quality. Several research are going on to restrict the copy control. IBM, Tokyo Research Laboratory initially proposed the utilization of watermarking innovation for DVD duplicate assurance in September 1996. The security and control can be kept up at the season of dispersing and distributing data, two methodologies might be utilized i) Use of copyright watermark, ii) Laying the foundation for digital right administration framework to get duplicate control of appropriated data where scholarly right assurance and duplicate control are significant concerns.

  5. v)

    Device management: - The device management watermarking may be a system within which, watermarks are constituted to manage access to an asset plus utilizing a corroboratory device. A few methods of watermarking are created in sound gadgets in past decades, including lowest-bit coding, spread range quantization. Video watermarking on a cell phone is an incredible test because of the constrained asset of the gadget. One of the methods for video watermarking in mobile transforms the video to YCbCr portrayal and embeds the watermark in the Y segment of the extracted frame of the video.

3 Features of Digital Watermarking

The features of watermarking plays an important role in the development of a watermarking system for the different applications[45, 77]. As there is a trade-off between these features, so one should keep in mind which feature is to be focused on. Various features of watermarking have been discussed in [2, 36, 87, 102] and is shown in Fig 2. Table 1 shows the vital characteristics and corresponding applications of digital watermark in short.

Fig. 2
figure 2

Feaures of Digital Watermarking

Full size image
Table 1 Application of watermarking corresponding to features
Full size table

Image fidelity: - Fidelity is the visual similarity between the watermarked image and its cover image. In other words, fidelity is the amount of imperceptibility of the watermark in the watermarked signal.

Robustness: if watermarks survive malicious (i.e. that destroys watermark) and non-malicious (do not explicitly mean to modify it) attacks then those watermarks are known as a robust watermark. Copyright protection is one of the applications were using this kind of digital watermark will be beneficial as they are more prone to malicious & non-malicious attackss.

Imperceptibility: The watermarked image ought to seem like the same as the cover image to the human eye. The spectator can’t identify that the watermark is inserted in it.

Security: An unauthorized user cannot distinguish, retrieve or modify the inserted watermark. Nowadays the researcher is giving prime importance to the security of watermark.

Effectiveness: This refers to the odds that the message in the watermarked picture would be accurately distinguished; In a perfect scenario, probability needs to be 1.

False positive rate: This refers to the number of digital works that are recognized to have a watermark inserted whereas in actuality it has no watermark inserted. Hence, it ought to be held less to watermarked frameworks.

Payload size: Data payload or payload size can be defined as the number of watermark bits embedded in a cover data.

Capacity: Without inherent redundancy presented by error-correcting codes for channel coding, the maximum repetition of data payload within the signal(audio, image. etc) is the watermark capacity.

Low Complexity: The cost of computation is directly related to thecomplexity of watermarking. As the complexity of watermarking will be increased automatically cost will also increase.

Verifiability: The watermark ought to have the capacity to give full and dependable proof to the responsibility for secured data items. It very well may be utilized to decide if the data is to be ensured and monitor the spread of the information being secured, recognize the validness, and control unlawful duplicating.

Security: Security in watermarking signifies the capability of resisting the intentional attacks. This feature is very vital to the watermarking system. The security requirements for watermarking schemes differ significantly from application to application.

4 State-of-the-Art

This section contains the recent research worked of many authors on different types of multimedia and database techniques. This section is further subdivided into the various subsections. Sections 4.1, 4.2, 4.3, 4.4, 4.5 and 4.6 deal with state-of-the-art of image, video, audio, text, graphics, and database watermarking techniques respectively. Limitations and challenges of watermarking techniques are discussed in section 4.7.

4.1 Images

In this section, we had discussed various recent state-of-the-art image-based on watermarking techniques like DCT, DWT, etc. on color, grayscale, medical, etc. images, which has been evaluated based on various performance matrices such as Bit Error Rate (BER), Peak-Signal-to-Noise-Ratio (PSNR), etc. Various attack like image processing and geometric attack, noise, filtering, tampering, etc. also has been performed to check the robustness, imperceptibility, etc. of these techniques for various applications like owner identification/verification, copyright protection, etc.

Su et al.[90] proposed a novel blind color image watermarking scheme. Here, QR decomposition techniques are used to embed the color image watermark into the color host image.

Su et al. [93] proposed a spatial domain blind color image watermarking technique for the copyright protection of color image. Here, by using algebra operation maximum Eigenvalue of Schur decomposition is obtained to embed the watermark.

Su et al. [92] proposed a novel watermarking scheme for the copyright protection of the color image using the DC coefficient and the AC coefficient of the color host image. Experiment results show that the proposed scheme shows better imperceptibility and also is robust against several attacks.

Su et al. [89] has proposed and investigated a spatial domain watermarking scheme for the copyright protection of the color image. Here, based on the features of the DC coefficient of DFT blind and the robust watermarking scheme is developed.

Su [88] has presented a novel blind watermarking based on Hessenberg decomposition. The presented schemes show lower computational complexity than other methods based on singular value decomposition or QR decomposition.

Qin et al [71] have proposed a completely unique self-implanting fragile watermarking technique for images. This system is developed for meddling recovery and is predicated on the reference-data interlocking mechanism and adaptation choice of embedding mode. Experimental results reveal this method is more powerful compared with the rumored schemes.

Roy et al. [78] proposed a method to avoid geometric and image transformation attacks. This method is independent of the cover image. It is purely based on the key. Experimental results reveal that this procedure works well for simple scenarios. It is robust. The fidelity of the cover is also maintained. However, in the case of complex attacks (cropping, rotation, scaling), this method falters. Comparative analysis suggests that the projected technique performs well each in terms of hardiness and time demand than several alternative watermarking techniques. Future work can emphasize rising the projected technique for incorporating higher resistance to compression attacks still as composite attacks involving a combination of scaling, cropping and rotation attacks along.

Parah et al. [66] planned a strong blind watermarking technique, supported block-based DCT constant modification. The experimental results show the common PSNR price of the projected scheme is 41.25 dB that is better than the number of state-of-the-art. Also, Normalized Correction (NC) value are higher than [91, 99].

Huynh et al.[30] planned a completely unique strong blind color image watermarking methodology. Hereby rotten a gray-scale image to binary pictures from LSB to MSB for the embedment, a gray-scale watermark is totally encoded into a color host image employing a quantization technique within the rippling domain. The experimental results prove that the planned methodology reaches a high performance in a physical property of embedded host pictures and hardiness of extracted watermarks and is superior to the remainder of state-of-the-art.

Aggarwal et al. [1] planned and compared four blind watermarking methods. Here the author has performed eight experiments for a close analysis of the pro- exhibit watermarking strategies S1, S2, S3, and S4. The experimentalresult shows that S3 and S4 have higher PSNR than [46, 56] whereas the S1, S2 has less PSNR.

Vaidya & PVSSR [100] planned a strong blind watermarking methodology using Bhattacharyya distance and mathematical function. The experimental results compared with the state-of-the-art and show a better result.

Thongkor et al. [97] presented a digital watermarking scheme for camera-captured images. In this scheme, to embed a binary image with the same size as the host image each pixel of the host image was used to carry a watermark bit. The average wPSNR of the proposed scheme is 35dB, whereas the average SSIM value is 0.93.

Lai [42] planned a method primarily based upon single value decomposition (SVD) & Tiny-Genetic algorithm. Here, the singular values of the cover image are adapted to embed the watermark. Simulation results have shown that the watermark that has been embedded is robust enough to resist attacks or image process operations and also the hardiness performance of this planned approach is superior to the opposite similar approaches.

Najih et al. [62] devised a contourlet transformation & quantization index modulation primarily based watermarking technique. Lagrange technique was also utilized for optimization. Experiments show higher transparency & more practical physical property &, additionally smart capability and provide better hardiness than other techniques for different attacks.

Sarreshtedari & Akhaee [84] proposed a way to handle image security. The idea was to encode the source channel. It was based on Reed- Solomon (RS) and set partitioning in hierarchical transforms (SPIHT). The initial segment includes the encoder- bits used for content recovery, the second section is made of parity bits and the last region comprises of check bits. The experimentresult shows the effectiveness and superiority of their proposed method in comparison to other methods.

The comparative analysis of state-of-the-art of image watermarking is depicted in Table 2, and Table 3 depicts the summary of image watermarking state-of-the-art.

Table 2 Comparative analaysis of image watermarking
Full size table
Table 3 Summary of recent state-of-the-art of image watermarking techniques
Full size table

4.2 Video

In this section we had discussed various recent state-of-the-art of video based on watermarking techniques like multi-resolution wavelet decomposition, chirp Z-transformation & entropy analysis, etc. Various attack like blurring, compression, change in brightness & contrast and geometric attack, noise, filtering and averaging, etc. also has been performed to check the robustness, imperceptibility of these techniquesfor various application like video copyright protection without dropping visual quality, cloud-assisted secure video transmission & sharing, secure blind video watermarking for medical purposes & to limit the pirated copy of digital video distribution, etc.

Venugopala et al. [101] proposed a bitstream video watermarking technique that is carried out by study of the temporal and spatial domain. In their experiment execution time of insertion & execution process was examined, using a mobile device. As a result of the experiment PSNR value of the extracted image is within the acceptable range, which was embedded in the video. Also, execution time and power devoured in a cell phone are inside satisfactory breaking points. Also, BER was constant for the entire image.

Preda & Vizireanu [70] has proposed a novel digital watermarking for video dependent on multi-resolution wavelet decomposition. They used the blind watermarking method & binary image as a watermark image. The watermark was inserted into the sub-bands i.e. LH, HL, HH by quantization. The performance was increased by inserting the same watermark in dissimilar frames of the video need improvement against geometric distortion, some detection problem and perceptual quality of watermarked videos & some arithmetical complexity.

Liu et al. [51] projected zero-watermarking novel sturdy techniques for 3- Dimensional videos for DRM- Digital Right Management based mostly upon TIRIs, 2D- DCT, VSS. In their experiment every depth maps and 2nd frames of input 3-D video are first of all smoothed and so subsampled at intervals spatial in addition as a temporal domain. Secondly, by averaging these frames & its depth maps TIRIs are generated. For 2D-DCT, transformations are performed on these TIRIs. Remaining low-frequency coefficients are then chosen for the extraction of coefficients of low frequency to form positive that the strength of content-based choices as a result of they embody the foremost energy of the initial TIRIs to an extent. First, the projected theme doesn’t insert the copyright data into the 3D videos and may avoid content distortion that is an improvement over the 2nd video frame-based scheme. More enhancements are needed on robustness against geometrical attackslike cropping & rotation.

Nezhadarya & Ward [63] proposed a semi-blind methodology to test the standard of video degraded by H.26/AVC decompression/compression. Using 2D spread transform modulation bits of watermark are inserted within the multiscale derivative domain. New merit-score (MS) was purposed to seek out the best watermarking parameter & to test the standard of assessment methodology. The results of the experiment show mistreatment STDM methodology yields higher benefit scores rather than inserting in one scale solely & high watermark capability and lustiness against distortion was obtained. Numerous alternative varieties of channel distortion weren’t tested like & packet loss and AWGN noise and another challenge is to estimate the standard in terms of other human sensory systems (HVS) based mostly video quality metrics, like VQM and MPQM.

Thanki et al. [96] proposed a hybrid watermarking scheme to achieve fragility and security (copyright ownership or authentication) scheme that uses curvelet transformation with combinations of DCT, CS (Compressive Sensing), DWT & SVD. The result of this scheme claimed faster execution time, high-payload capacity & multimedia data authentication. The quality measures values of the projected theme also are higher than quality measures values of existed schemes within the literature. Further to improve performance instead of wavelet transform, curvelet transform can be used & real-time implementation yet need to perform.

Rasti et al. [73] proposed imperceptible non-blind & robust video frame water-marking technique using QR decomposition SVD; Chirp Z-transform (CZT), DWT & entropy analysis. To evaluate the robustness Correlation Coefficient (CC) metric (variant to contrast & brightness) is used. In this method, frames are divided into moving and non-moving parts. The block-based watermarking theme was used for non- moving parts of every color channel. More for embedding the watermark image instead of the common entropy of all blocks, blocks with entropy lower are subjected. When witnessing the experiment with common signal process attacks results show the proposed scheme is strong & impalpable in nature & performs better than its state-of-the-art.

Loganthan & Kaliyaperumal [55] proposed a reversible adaptive video watermarking technique based on neural network, fuzzy inference system (HVS based) & bidirectional associative memory (BAM). The first BAM neural network trains multiple watermarks. Second fuzzy inference system of HVS base, embedding adaptive factors are computed that is used to infix the weight matrix generated on the lower video end in its both components luminance & chrominance. Using PLA, coefficients embedding is done. The result of this experiment shows this technique is better where a large number of watermarks are required with good imperceptibility, robustness & good watermark embedded capacity in comparison to other methods. The experiment fails to withstand medium filter Gaussian filter & rotation attacks. Also, it doesn’t support fast motion video.

Hossain et al. [26] in order to secure video sharing & transmission proposed a cloud-assisted framework, where mobile client’s capabilities are limited. In their proposed work keyframes are deleted using GA when a video is captured by smartphones. Watermark is inserted in the keyframes using the DWT based watermarking technique. Based on error-correcting codes a two-way layer protection mechanism is applied to the identity or signature of an individual to create a watermark that is embedded into the video to protect against attack & transmission loss. Using Shamir’s secret theory key is shared among entrusted entities. Then watermark is transmitted to the cloud. SSIM and PSNR were used to carry imperceptibility analysis of the video. The proposed framework has space optimum transmission and security.

Cediclo-Herandez et al. [11] have worked on to improve the video watermarking schemes where distortion is below the sensitivity threshold. They proposed a profile, Saliency- modulated JND (Just Noticeable Distortion) that adopts watermark strength to obtain imperceptibility & robustness. First JND pro le is created using 3 steps (i) Saliency mapping, (ii) JND estimation, (iii) modulation stage. Second JND is used for tuning a basic video watermarking technique’s energy. As a result, this experiment generates a lot of powerful video watermarking methodology with a gain of 14.6dB approx. where an unmodulated JND profile gain is 3.11dB. Model V of their approach obtains the most effective performance when put next to different models.

Madine et al. [57] presented a robust blind watermarking scheme for the raw video signal. Here, the watermark is embedded into the HH subband coefficient of the 3 level 2D-DWT of the video frame. The proposed scheme offers low computational complexity which eases the implementation.

Yassin et al. [108] introduced a blind digital watermarking video theme. For security, one secret key’s used throughout the recovery of the watermark. In their experiment, using DWT, every video frame is decomposed into a variety of sub-bands. To Quantize the most constant blocks of PCA of every sub-band, quantization Index Modulation (QIM) is employed. Results reveal high imperceptible property. The experiment was conducted on two medical videos. This scheme was extremely strong against several attacks like histograms equalization, noise, gamma correction, JPEG coding.

Nouioua et al. [65] conferred novel and powerful digital video watermarking technique using the SVD. During this recommended technique, the author has resolved the matter of embedding the watermark. It’s been done by choosing solely the frames that have huge motion energy that is appropriate to the HSV. The comparison of the results with different video watermarking techniques indicates the prevalence of their theme.

Farri & Ayubi [17] has given a strong secure and video watermarking technique supported whole number rippling remodel and also the generalized chaotic trigonometric function map. The experimental results show that the PSNR value of the projected techniques is 45.07 dB.

Asikuzzaman et al. [10] has planned a basic blind digital video watermarking scheme based on the DT CWT. In the planned technique, the author has embedded the watermark into the low-frequency parts of the U channel in a YUV illustration wherever these components guarantee hardiness and exploitation the U channel enhances the physical property. The experimental results show that this theme is additionally sturdy to compression, cam-cording, watermark estimation re-modulation, temporal frame averaging, multiple watermark embedding, different geometric attacks and downscaling in resolution

The comparative analysis of state-of-the-art video watermarking is depicted in Table 4, and Table 5 depicts the summary of video watermarking state-of-the-art.

Table 4 Comparative Analysis of Video Watermarking Techniques
Full size table
Table 5 Summary of recent state-of-the-art of video watermarking techniques
Full size table

4.3 Audio

In this section, we have discussed various recent state-of-the-art audio based on watermarking techniques like DWT, DCT, Arnold Transformation, Entropy, Fast Fourier Transform (FFT) spectrum, etc., ona various set of audio clips of different length. This has been evaluated based on various performance metrics such as signal-to-noise ratio (SNR), Segmental Signal-to-Noise Ratio (SSNR), Mean Opinion Score (MOS), BER, Percent Root-Mean-Square Difference (PRD), and Time-Scale. Modification (TSM), Objective Difference Grade (ODF) & NC etc. Various attacks like AWGN, re-sampling, re-quantization, amplification, cropping, noise, filtering, echo, jittering, stir-mark also has been performed to check the robustness for various applications like authenticity verification of audio signals, copyright protection & privacy protection in biomedical signals, etc.

Saadi et al. [82] proposed a method of blind watermarking of audio signals and speech. After the signal is framed, they used the DWT and then applied the discrete cosine transform (DCT) on each frame. For correlation purposes, the frame is decomposed into two segments to perform sub-sampling. In order to a security concern, Arnold transform is applied to the watermark. Without using the insertion parameter & original speech/audio signal the fully blind detection is accomplished. Experimental comparisons and assessments of their scheme with other schemes determine a good balance between robustness, security, capacity & imperceptibility. The decomposing with sub-sampling declines robustness against the re-sampling attack.

Hu et al. [27] introduced a blind watermarking scheme in audio files using distributive characteristics of the wavelet coefficient. Results establish the strength of the projected LWT-SSR against time-shifting and time-scaling attacks and customary signal process operations compared with four different progressive techniques.

Renza & Lemus et al. [74] introduced a new fragile scheme for audio forensics purposes, like digital audio authenticity based on the OVSF (Orthogonal Variable Spreading Factor) & QIM. The main feature of their proposal is that the process which is embedded is accustomed in accordance with the amplitude/length and the value/length of the mark of the audio signal. Through quantization index modulation (QIM) in the wavelet domain using a min value of quantization support the embedding process which increases the fragility. Kappa index, sensitivity, and specificity used for performance analysis against several attacks.

Hwang et al. [31] has proposed QIM-based watermarking techniques for stereo audio signals that fully exploits the key features of SVD. As the proposed method efficiently exploits the ratio of singular values, the embedded watermark is extremely imperceptible and robust against volumetric scaling attacks.

Dhar & Shimamura [14] has proposed an audio watermarking scheme using Log-Polar Transformation (LPT) and entropy-based on SVD in DCT domain. First, their scheme utilizes entropy, LPT, DCT, quantization, and SVD jointly. In the end, the highest entropy DCT sub-band is used to obtain the highest singular value and its Cartesian component is quantized to embed the watermark. Simulation results prove that this technique can be used for the purpose of copyright protection of the audio signal.

Fallahpour & Megias [16] proposed an audio watermarking system with high- capacity to insert and extract data way by changing the certain number of FFT spectrum magnitudes by Fibonacci number. A particular frequency band of the FFT spectrum was selected to insert secret bits. Secondly, a large Fibonacci number (lower than the magnitude of each FFT spectrum) was calculated. These were then embedded in each frame. Results reveal the algorithm to be robust with a capacity of 700 bps to 3 kbps. The comparison proves the prevalence in each capability and imperceptibility of the recommended technique with relevance to different techniques within the literature.

Xiang et al. [105] suggested an audio watermarking method based upon orthogonal PN (Pseudo Noise) sequence, DCT, variable embedding strengths and polarities while preserving the embedding capability. During embedding, DCT is applied to the audio signals to obtain segments of audio in the DCT domain. Post this, a bunch of orthogonal PN sequences is generated, each containing several bits for watermarking. An embedding algorithm is then used to introduce a watermarking sequence in an audio sample. The watermarked bits are obtained by comparing & computing the correlations among orthonormal PN sequences and the watermarked audio segments. Some of the pros of this scheme include robustness, high embedding capability, quality preservation & low computational complexity. Simulation results demonstrate the superior performance than the other progressive technique.

Hua et al. [28] with optimized imperceptibility and hardiness, they proposed an audio watermarking methodology that supported time-spread echo. An FIR- filter based on convex optimization was accustomed to getting the best echo filter coefficients. The echo filter power spectrum is molded by (ATH- Absolute Threshold of Hearing) & (MPSM-Maximum Power Spectral Margin). However, there was scope for improvement against de-synchronization attacks. Though relaxation has been employed in the improvement for economical solutions, the designed watermark still enjoys vital improvement in terms of each imperceptibility and hardiness as compared to the present state-of-the-art solutions [105] and [106].

Xiang et al. [104] proposed a patchwork-based audio watermarking strategy against de-synchronization attacks such as jitter attacks, time-scaling and pitch-scaling utilizing DCT and logarithmic DCT (LDCT). The proposed scheme shows high embedding capacity compared with other audio watermarking techniques.

Lei et al. [44] proposed a Quaternion Wavelet Transform (QWT), Self-Adaptive Particle Swarm Optimization (SAPSO) and chaotic outline based audio watermarking scheme. The highlight of this schemes are - (i) an ideal adjust of the conflicting watermarking prerequisites is decided by the SAPSO calculation without SAPSO parameters tuning; (ii) both SVD & QWD are investigated to improve performance; and (iii) a (MSS - Modified Spread Spectrum) based watermarking strategy is proposed to embed the watermark bit & synchronization code utilizing the 4D QWT coefficients. The proposed algorithm is exceptionally robust against attacks like re-quantization, resampling, MP3 compression, additive noise without significantly corrupting imperceptibility. The method comes about to illustrate that the proposed algorithm outflanks SVD- related, optimization-based, and conventional wavelet techniques. It is seen that their projected scheme outperforms the present chosen audio watermarking Schemes with respect to SNR and Corr values.

Yuan et al. [110] proposed a novel advanced audio watermarking technique based upon strong (DT CWT - Dual-Tree Complex Wavelet Transform) and (MFC- CFD - Mel-frequency Cepstral coefficients feature detection). The vigorous MFC- CFD strategy is proposed to extricate the highlight segments that ought to be migrated when they have audio signal attacked by different mutilations counting both the common & conventional geometric mutilations. The direct relationship is calculated to evaluate the presence of the watermark amid the watermark detection. Experiments reveal that the suggested methodology can accomplish robustness against MP3 compression, low-pass filtering, normalization, volume alter, geometric distortions, like pitch invariant TSM, resample Time-Scale Modification (TSM) and beat invariant pitch moving. The comparison results reveal that the projected scheme performs far better than its state-of-the-art.

Ali et al. [7] projected a zero-watermarking for the privacy protected healthcare system. Here the proposed framework includes 2 modules, initial zero- watermarking to secure the identity of a person. The second module is made for the invention of vocal overlay muddle. The performance of this approach is assessed by utilizing the MEEI voice clutter database. The experimental results show that the projected algorithmic rule is reliable within the detection of a subject’s identity and strong against noise attacks with varied SNR when putting next to alternative state-of-the-art technique.

Table 6 depicts a summary of the state-of-the-art audio watermarking techniques.

Table 6 Summary of recent state-of-the-art of audio watermarking techniques
Full size table

4.4 Text

In this section, we have discussed various recent state-of-the-art text-based watermarking techniques like character encoding & attributes, open word space, homoglyphs substitution, semantic role labeling, new-defined character, etc. Various attacks like deletion, insertion copying, pasting, replacing, etc. also been discussed.

Rui et al. [81] suggested a watermarking method based upon character encoding and properties for texts blended in Chinese and English. To begin with, information required to watermark was encoded after that key was connected and after that, it was stratified. The MD5 encryption technique is used to enhance security. This scheme shows a high embedding capacity.

Alotaibi & Elrefaei [8] proposed two imperceptible watermarking methods for Arabic content based on the pseudo-space. Within the first proposed strategy, based on dotting include in Arabic content the pseudo-space is embedded before and after typical word space. The second proposed strategy increment the capacity by embedding the pseudo-space and extra three small or zero-width spaces, where the absence demonstrates bit “0” and the presence of them shows bit “1”. Utilizing variable measure text samples with different watermark lengths for the proposed strategy a few tests are conducted. As a result of this, Strategy 2 has the most noteworthy capacity but marginally lower imperceptibility than Strategy 1. Also, they are vigorous against electronic content attacks such as content designing, replicating and sticking, and content altering for altering proportion up to 84%. Further, rather than utilizing binary bits text watermark can be utilized with compression calculations such as Huffman calculation to represent the watermark. Within the proposed watermarking strategies, a private key may well be utilized to demonstrate the genuineness. The capacity results of the proposed methods are higher than the other watermarking scheme.

Mir [60] has proposed a web-based text watermarking scheme for HTML text in the webpage. HTML body is parsed to serve as the watermark. Post parsing, a hash function is used to embed undetectable control characters. This results in an undetectable set of watermarks. Specific verbs (is, are), articles (a, an), & often occurring prefix letters (wh, th) of English dictionary are used.

Rizzo et al. [75] suggested a watermarking method able to work on all the SM (Social Media) stages based on the homoglyphs substitution. Hash work is used to embed the text content by substituting their homoglyphs with unique symbols. For evaluation, they select 18 different SM stages, utilizing 6,000 posts from 6 profiles of public personalities.

Halvani et al. [22] proposed four natural language watermark embedding scheme, to operate on the lexical and syntactic layer of German texts. The presented scheme has the advantages like sovereignty from complex NLP methods and rich lexical resources. Also, the proposed scheme shows better run-time efficiency.

Liu et al. [50] proposed strategies for graphic digital text Watermarking. This comprises 8 levels, etymon, line, push, section, page, character, pixel, and chapter; 3 parts, structural feature, similitude, and self-characteristics. They portray the implementation and focuses on the effectiveness of hypothetically.

Chen et al. [12] proposed a new technique for watermarking text which employs the linguistics roles to implant the watermark information. (NLP- Natural Language Processing) is applied to seek out and tag the 3 types of linguistics elements A0, A1, and ADV in content. Here, the watermark is changed into the hexadecimal Unicode and then compressed using Huffman encoding.

Zhang et al. [111] proposed a modern watermarking algorithm for Microsoft Word reports based on newly defined characters utilizing TrueType. The proposed algorithm specially designed for the copyright assurance of Microsoft Word records. In this scheme, the TrueType function produces newly defined characters that are utilized as a watermark, and after that concurring, to a few rules, this watermark is implanted into the MSWord document. This algorithm was simulated using C++. Results show that this algorithm has high strength, total imperceptibility against numerous sorts of attacks and can find the alter range. Additionally, it can be connected to both English and Chinese. In spite of the fact that the scheme has effective performance, it can be optimized, for illustration, the semantic approach can be utilized to decide the inserting positions, and the watermark can be prepared before implanting, etc.

Zhang et al. [112] presented a watermarking procedure for texts written in Chinese. The methodology inserts the watermarking signals into a few Chinese characters by modifying the dimensions of closed rectangular regions in these elements, thus it’s entirely primarily based upon the content. Compared to strategy [94], this is often a lot of economical in concealing the characters.

Al-Sewad et al. [6] planned a text/content watermarking scheme for making certain the possession and property rights of the color images. Inserting is accomplished by embedding texts indiscriminately into the color image as clamor. They embraced the YIQ image for the implanting method as a result of it had been found to be faster than different image handling ways. A discretionary alternative of encrypting the text watermark before inserting is in addition prescribed, where, the content may be encrypted utilizing any enciphering procedure.

Summary of the recent state-of-the-art text watermarking is depicted in Table 7.

Table 7 Summary of recent state-of-the-art of Text watermarking techniques
Full size table

4.5 Graphics

In this section, we had discussed various recent state-of-the-art graphics based on watermarking techniques.

Doncel et al. [15] presented an optimal blind detector structure for watermarked polygonal lines in 2 D vector graphics data. Here, Detection error probabilities and ROC curves are used for the performance analysis of the proposed detector.

Lin et al. [49] proposed semi-blind and semi-fragile reversible watermarking techniques for authenticating 2D vector graphics. Here, for the authentication purpose principle of bionic spider web is used.

Xio et al.[107] has proposed combined reversible watermarking techniques for 2D CAD engineering graphics. This scheme has been developed by utilizing the concept of Improved Quantization Index Modulation (IQIM) and Improved Difference Expansion (IDE).

Peng et al. [67] proposed a reversible watermarking scheme based on iterative embedding with virtual coordinates. Experimental results shows that, this scheme can be applied for content authentication and secret communication in 2D CAD engineering graphics.

Summary of the recent state-of-the-art graphics watermarking techniques is depicted in Table 8.

Table 8 Summary of recent state-of-the-art of graphics watermarking techniques
Full size table

4.6 Database

In this section, we had discussed various recent state-of-the-art of database based watermarking techniques like twice-embedding method, etc. on various data sheets, RDBMS Tuples, etc. which has been evaluated based on various performance matrices such as numerical certainty level, PSNR, Alternation ratio, similarity score, false hits & miss rate & NC, etc. Various attacks like substitution, addition, alteration, vertical partition, invertibility & Mix-Match also has been performed to check the robustness for various applications like database copyright protection, protect valuable numerical relational data from illegal duplication and redistribution.

Guo et al.[20] suggested a twice-embedding approach that uses a fingerprinting solution that acts as a guard to valuable numeric relational data against illegal replications and redistribution. In the principal embedding method, they embed a fingerprint that is unique to recognize every beneficiary. The embedding technique is helmed by a secret key. The second implanting method intends to confirm the extricated fingerprint and give a numerical certainty level. The results illustrate that their arrangement is strong and practical to distinctive attacks.

Zhou et al. [113] proposed a strategy named WDI (Watermarking Databases utilizing Image) i.e. BMP Bit Map Image embedding in the RDBMS to guarantee information’s copyrights. The robustness of the algorithm was progressed utilizing BCH (Bose Chaudhuri-Hocquenhem) coding. Moreover, a Trusted Third Party (TTP), is used to embed and recognize the watermark. Besides, they look at the flexibility of the algorithm hypothetically depends upon the principles of insights in detail. Tests illustrated the strategy proposed is robust to various sorts of attacks with the objective that the copyrights can be effectively guaranteed.

Gross-Amblard [19] suggested a Query-preserving technique applicable to databases and XML pages. They first saw that unrestricted databases cannot be watermarked while protecting trivial parametric queries. Here, theauthor has proved that watermarking on the arbitrary instances is not possible.

Unnikrishnan & Pramod[98] suggested a method that relies on a hybrid algorithm HOLPSOFA for relational databases. HOLPSOFA is a mix of Orthogonal Learning Particle Swarm Optimization (OLPSO) & Fiery Algorithm (FA). This methodology joins the benefits of FA & OLPSO, which can discover the optimal-time results at the same time. The relational database watermarking method comprises three phases, (i) Optimal area ID through1 HOLPSOFA algorithm (ii) Watermark inserting and (iii) watermark extraction. They also compared the HOLPSOFA algorithm with OLPSO and FA. NC and MSE are used for the performance analysis of watermarking. This is a robust strategy and can withstand different types of attacks like insertion, alteration deletion, etc.

Gupta & Pieprzyk [21] has presented a blind and reversible watermarking model. The capacity of the presented watermarking technique is high and is having the attack resistance probability in between 89 and 98 percent. Here, to achieve reversibility authors utilize difference expansion on integers.

Perez Gort [68] devised a watermarking technique based on the AHK algorithm for the database to increase the embedding capacity of the watermark. Here, Attribute Fraction is used to reduce the number of marked tuples. The author has improved the embedding capacity with this scheme.

Pournaghshband [69] has proposed an effective watermarking framework for relational information that’s vigorous against different attacks. Whereas past strategies have stressed around bringing errors into the real data, this technique inserts unused tuples (“fake” tuples), to the relation and it acts as watermarks. In comparison with the previous approaches presented insertion algorithm is probabilistic and the detection algorithm is somewhat deterministic.

Agrawal et al. [3] projected a watermarking methodology for relational data. This strategy ensures that some bit places of a few of the properties of a parcel of the tuples embrace specific values. Explicit bit places and values are chosen algorithmically under the influence of a secret key. This watermarking strategy has four essential tunable parameters. Authors, using DB2 showed that the presented scheme can be used for real-time applications.

Khanduja [35] in this work has focused on security analysis of the work done in the field of database watermarking techniques. Here author has categorized the watermarking systems into four kinds, (i) ATSASB (all tuples, single attribute and single bit), (ii) MTSASB (multiple tuples, single attribute and single bit), (iii) MT- SAMB (multiple tuples, single attribute, and multiple bits) and (iv) MTMAMB (multiple tuples, multiple attributes, and multiple bits. The author has analyzed the security of the watermarking scheme hypothetically investigated its reliance on different parameters :(i) Nr, (ii) Mt, (iii) Lpa, (iv) Npa.

Summary of the recent state-of-the-art Database watermarking techniques is depicted in Table 9.

Table 9 Summary of recent state-of-the-art of Database watermarking techniques
Full size table

4.7 Limitation and Challenges of Watermarking

In the last two decades, lots of work has been done in the field of watermarking. But still, there are various limitations and challenges in the development of watermarking techniques. In this section, the various limitations and challenges of watermarking are discussed.

Robustness, imperceptibility, payload, and computational cost are the major features of watermarking. There is always a trade-off between these features. It is practically impossible to design a watermarking system to address all these watermarking features[95]. Also, it is practically not possible to develop a distinct system robust to all the well-known attacks at the same time [29, 95]. The development of watermarking scheme to address satisfy these tradeoff is the main challenge [61]. The reversible watermarking field lacks benchmarking tools for its evaluation[34]. Also, there are no industry-wide standards for watermarking in the DRM application[33]. Watermark based forensic techniques had alimitation that watermarks need to be embedded in the multimedia before distribution. The spatial domain watermarking scheme is commonly used for authentication. But, poor robustness against the various attacks is its major drawback[30, 100, 108]. A block-based watermarking scheme is having disadvantages that it is unable to embed a watermark in all the blocks, which leads to low capacity. Recently, several works have been done in the frequency domain. But the high computational cost and low payload capacity is its major limitation [65, 89, 96]. Also, the DWT watermarking suffers from three major drawbacks (i.e. poor directional information, shift sensitivity, and lack of phase information), whereas false-positive problems and higher computational cost are the main drawbacks of the SVD-based watermarking[72]. The selection of optimal scaling factors is the major challenge in DWT as well as SVD based watermarking techniques[24, 100, 104]. Also, the DWT based scheme suffers from rotation attacks. Payload or capacity is the main limitation of the video watermarking scheme. In the visible watermarking scheme, the watermark is visible in the watermarked signal (image, video, and graphics). Visible watermarking can be effortlessly tampered by the attackers with the use of image processing mechanisms [18]. For a 2- D CAD engineering graphics IQIM based scheme is commonly used. The major drawback of the IQIM based scheme is that not every vertex is embeddable[107]. For audio, SS based watermarking is commonly used. A drawback of this technique is the host signal interference problem[105]. This drawback could significantly lower the robustness of watermark extraction. The robustness of audio watermarking against recapturing and desynchronization is still a challenging issue[53]. The study reveals that the main constraint of the existing audio watermarking methods is the difficulty to achieve a favorable trade-off among imperceptibility, robustness, and data payload[14]. Watermarking text on Social Media is having alimitation and specific requirements[75]. The watermark should preserve the length as well as the content of the original text without converting the text into the images. Watermarking natural language is still a challenge in the domain of digital watermarking. Multi-attribute techniques, a common limitation is that often they define a fixed number of attributes for embedding the marks[68]. Large volume and redundant data is the major challenge for database watermarking [35]. Difference expansion based watermarking techniques is one of the major watermarking technique for database and is unable to increase the capacity of the relation without distortion tolerance.

5 Performance Measures in Digital Watermarking

Performance metrics are required for calculating or verifying the effectiveness of watermarking techniques[41, 43, 64]. Some of them are given below:-

Mean Squared Error (MSE)

is used to verify mutilations between cover image & watermarked image. This helps to recognize any alteration within the watermarked image.

$$ M.S.E=\frac{1}{n}\sum\limits_{i=1}^{n}(A_{i}-A_{i}^{*})^{2} $$
(1)

Here A signifies the cover image and A* signifies the watermarked data.

Euclidean distance (ED)

In a Euclidean space, it is the common distance between two points. Two-dimensional Euclidean distance is utilized for images. The Euclidean distance between two images is given by: -

$$ ED(A,A')=\sum\limits_{i=1}^{M}\sum\limits_{j=1}^{N}(A_{(i,j)}-A'_{(i,j)})^{2} $$
(2)

Peak-Signal-to-Noise Ratio

PSNR utilizes mean squared error to check bending between the watermarked and cover image. PSNR can be calculated by the below mentioned formula. It is widely used to investigate reformation of lossy images. Image is the information here while noise is the error.

$$ PSNR=10log_{10}\left( \frac{MA{X_{i}^{2}}}{MSE}\right) $$
(3)

Normalized Correction

Normalized correction is a measure of the degree of similitude of two images as a function of a time-lag connected to one of them. The following equation is utilized to calculate normalized correction for two images.

$$ NC(A,A')={\sum\limits_{i}^{M}}{\sum\limits_{j}^{N}}\frac{A_{(i,j)}*A'_{(i,j)}}{{{\sum}_{i}^{M}}{{\sum}_{j}^{N}}(A_{(i,j)})^{2}} $$
(4)

Hamming distance (HD)

is applicable for binary images. It can be utilized to measure the exactness of recouped watermark quantitatively. There equations are given underneath. Here, B= original image & B’ = processed image, M = width of the image & N = height of the image B (i, j) = pixel position at (i, j) location of B & B’ (i, j) = pixel position at (i, j) location of Y’.

$$ HD (B, B') =\sum\limits_i^M\sum\limits_j^N|B_{(i,j)}-B'_{(i,j)}| $$
(5)

Bit error rate (BER)

BER is used for binary images. It is utilized to measure the exactness of recuperated watermark quantitatively. There equations are given underneath. Here, B = original image & B’ = processed image, M = width of the image & N = height of the image B (i, j) = pixel position at (i, j) location of B & B’ (i, j) = pixel position at (i, j) location of Y’.

$$ BER (B,B')=\frac{HD(B,B')}{M*N}*100\% $$
(6)

Image Fidelity (IF)

IF decides the likeness between the watermarked and un-watermarked image. Higher the IF, the more imperceptible the implanted information is within the watermarked picture.

Bit correction rate (BCR)

It is also used for binary images. It can be utilized to degree the precision of recouped watermark quantitatively. There equations are given underneath. Here, B = original image & B’ = processed image, M = width of the image & N = height of the image B (i, j) = pixel position at (i, j) location of B & B’ (i, j) = pixel position at (i, j) location of B’.

$$ BCR (B,B')=\left( 1-\frac{HD(B,B')}{M*N}\right)*100\% $$
(7)

Structural Similarity Index (SSIM)

SSIM is used to measure the similitude between two images on the basis of their structure. The basic discernment is made based on pixels interdependence with its neighboring pixels. It is better than PSNR & MSE. Neighboring pixels contain critical data in regards to the structural composition of the image.

$$ SSIM (a,b)=\frac{(2\mu_{a}\mu_{b}+c_{1})(2\sigma_{ab}+c_{2})}{({\mu_{a}^{2}}+{\mu_{b}^{2}}+c_{1})({\sigma_{a}^{2}}+{\sigma_{b}^{2}}+c_{2})} $$
(8)

μathe average of x & μb the average of y. \({\sigma _{a}^{2}}\)the variance of a & \({\sigma _{b}^{2}}\) the variance of b, \(\sigma _{ab}^{2}\) the covariance of a and b. c1 = (k1L)2, c2(k2L)2 two variables to stabilize the division with weak denominator

6 Attacks on Watermarking.

One of the significant characteristics of any watermarking scheme is its robustness against attack. With regard to watermarking, an attack is anything that will harm or debilitate the discovery of the watermark. The processed watermarked information is at that point named as attacked information. In spite of the fact that accomplishing robustness against attacks remains an open issue, a few strategies can survive worldwide attacks. Effective strategies, for the most part, depend on i) embedding data in a space that’s invariant to geometric attacks, ii) utilizing layouts, iii) utilizing self-synchronizing watermarks, or iv) utilizing highlight points to realize synchronization.

Watermark attack can be broadly classified into 4 categories namely, removal, geometry, protocol, and cryptographic[39, 86, 103]. These are again subdivided into other sub-classes (Fig. 3).

Fig. 3
figure 3

Attacks on watermarking

Full size image

6.1 Removal Attack

Removal attack aims at the total expulsion of the watermark data without breaking the security of the watermarking algorithm. Advanced removal attacks attempt to optimize operations like quantization or de-noising to impede the inserted watermark as much as conceivable while preserving the quality of the attacked document. Removal attacks can be divided into 4 categories namely Sharpen, Blur, Median Filter, & Noise. The noise can further be classified as, Gaussian noise, Salt & pepper noise, Poisson noise, and Speckle noise.

6.2 Geometry Attack

Geometry attack does not essentially apportion with the watermark itself but it mutilates the watermark locator synchronization with the embedded data. Each geometric attack is characterized by a set of parameters that regulates the operation on the target. Geometry attack can be gathered into 4 classes specifically Rotation, Scaling, Translation, and Cropping.

6.3 Protocol Attack

In a protocol attack, the attacker adds his own watermark to the host data. Such attacks pose a threat to modern digital systems. Examples include the replication of a valid watermark to name a few. Protocol attacks are of 2 types – Invertible & Copy Attack.

6.4 Cryptographic Attack

Cryptographic attacks aim to break the security provided by watermarking. They may be of two types namely Oracle & Brute-Force.

Oracle attack: Oracle attack was first introduced by Vaude-nay at Euro Crypt ’02. In this, the attack works under the impression of an Oracle which on an encounter of a ciphertext, decrypts it and sends a valid or invalid reply to the sender. It assumes that the attackers can retrieve the padded messages encrypted in CBC mode and have access to the padding oracle. As a result, the attacker can recuperate the plaintext in regard to the cipher content utilizing approx.128b oracle calls, where b suggests a number of bytes in a block. Brute force attack: In this attack, the impersonator tries all possible keys to decrypt the message. It is also known as an exhaustive key search. The amount of time needed to decipher a cipher is proportional to the secret key size. Another kind of attack is a dictionary attack where the impersonator tries to make a guess of the password by using popular expressions or existing words.

7 Conclusion & Future Direction

In this work, we have presented a brief overview of watermarking systems. We have also presented the detailed classification of the watermarking which has been done by the various researcher. Further, this paper presents the various features of the watermarking approach followed by the summary of various state-of-the-art watermarking approaches and the various attacks on watermarking.

Based on the extensive discussion in the previous sections, it has been found that there is a trade-off between the various features of watermarking. In a single watermarking scheme it is very difficult to address these trade-off issues. The researcher can work to develop a watermarking technique to address these trade-offs. Also, the reversible watermarking field lacks benchmarking tools for its evaluation. The researcher can work for the development of a benchmarking tool to evaluate the reversible watermarking techniques. In image watermarking, the possible research area includes, real-time implementation of watermark, blind watermark detector, better perceptual model, and dual watermarking techniques. In image watermarking, the security of the watermark is given less priority by the researcher in comparison to other features. So, the researcher can also work on the improvement of watermark security. In the area of video watermarking, theresearcher should work to shorten the operation time and to meet the real-time requirements. At the present time, 3D printing models are becoming more popular and commonly used in applications such as medicine, manufacturing, architecture, end-user parts, product development, etc. Also, very few watermarking schemes is available for the animation so, the researcher may work in this area. In, audio watermarking, attacks such as TSM and cropping are prominent challenges for theresearcher. Lots of work, is needed to overcome this challenge. In database watermarking scheme computational time and robustness zero distortion in original data is the prime concern. Text watermarking scheme is generally applied to a particular alphabets only. This diminishes the usability and suitability of the text watermarking scheme. Text watermarking should be applicable to any kind of text. The researcher can identify and proposed an adequate scheme to resolve these issues. Authors believe that this survey paper is helpful for the researcher to work in the direction of data authentication, security and copyright protection of multimedia and databases.