A Secured Frame Selection Based Video Watermarking Technique to Address Quality Loss of Data: Combining Graph Based Transform, Singular Valued Decomposition, and Hyperchaotic Encryption

)e advancement of Internet technologies has led to the availability of audios, images, and videos in different forms. )e unauthorized users are exploiting the use of multimedia by transmitting them on various Internet sites to earn money unethically without the intervention of the original copyright holder. Watermarking is a technique used to hide the signal known as watermark inside multimedia data that is not visible to the intruder to manipulate any information. In this paper, a secured watermarking approach is developed to tackle issues related to copyright protection and ownership identification. A Secured Graph Based Transform, Singular Valued Decomposition, and Hyperchaotic Encryption hybrid techniques are proposed. )e watermark cannot be embedded in every frame of the video as it adds to the size of the video and watermark can be easily retrieved by an intruder. )erefore, the frame selection algorithm has been proposed in the given work. Adding watermark in the frame adds to the challenge of quality loss. )e quality loss is addressed in this work. Various attacks have been applied on the watermarked frames to calculate the performance of the proposed technique using quality metrics: Peak Signal to Noise Ratio, Structural Similarity Index, Normalized Correlation, and Bit Error Rate. )e results indicate that the proposed technique is effective against various attack scenarios.


Introduction
e availability of multimedia data across Internet has prompted unauthorized persons to illegally distribute multimedia data such as videos across the Internet. e issues like copyright protection and ownership identification are prominent and the development of the secured technique is required to counter these issues. Videos are the most attackable multimedia data and unauthorized people are distributing videos for their own benefits and are earning lots of money in this regard. e illegal distribution of video is illustrated in Figure 1 where videos are exposed to the Internet after DVD release or movie release and this problem has led to huge loss of movie industry. e real time videos are gaining lots of popularity with various OTT platforms like NETFLIX and AMAZON PRIME. e problem of copyright protection again emerges as the videos from these platforms are getting released to the Internet and thus drops the number of users accessing these websites. ere is a need of a secured technique to identify these unauthorized users, thus stopping this illegal distribution. Watermarking is a technique that embeds secret and unnoticeable signal inside the video which is unidentifiable to any unauthorized user. e watermarking embeds encrypted watermark inside the multimedia data and the process of extraction is done from the researcher's side to test validity of scheme. e videos need to be watermarked before they are distributed across the network. e major challenges in embedding the watermark in the video are the selection of frames and quality loss after embedding. e selection of frames is done because watermark cannot be embedded in every frame of the video as it will be very easy for the unauthenticated user to detect watermark and remove it. e quality loss is the major constraint in the research as embedding of watermark affects quality of the video. ere are many watermarking techniques available. e different types of techniques are described in Figure 3. e most common types of watermarking techniques are frequency domain techniques such as Discrete Cosine Transform (DCT) and Discrete Wavelet Transform (DWT). Many researchers are applying these techniques to provide solution to these problems. Although these techniques are good, improvement can be done in many aspects. e encryption mechanism used in the proposed technique adds to additional security feature as it would be very difficult for any intruder to detect watermark and recover watermark from watermarked video. e proposed technique used in the research is based on Graph Based Transform along with Singular Valued Decomposition. is combination is applied to encrypted watermark. e encryption can be done in many ways such as Ciphers, AES, and DES. e reason why AES and DES are not used for encrypting watermark is that they are very compressed and make the watermarking algorithm even more complex so the hyperchaotic encryption [1]  e organization of this paper is as follows: Section 2 reviews all the related work done in this area, Section 3 presents the research methodology of the proposed technique, Section 4 presents results gathered from various experiments performed on selected set of videos, and Section 5 describes conclusion obtained from the given research.

Related Work
e number of video watermarking techniques has been proposed in the field of watermarking. e most prominent watermark embedding technique is Discrete Wavelet Transform (DWT). is technique has been applied by many researchers as transformation of a frame to DWT is a reliable method. Spatial Domain Methods given in this paper are fast but not robust enough to handle any signal processing attacks [2,3]. Frequency domain techniques like DWT and DCT have been used by many researchers. e techniques are good but suffer from the problem of dimensionality reduction; that is why these techniques were coupled with another technique named as Singular Valued Decomposition (SVD). e major constraint in watermarking is the area where watermark is embedded. Many techniques and methods have been proposed nowadays where study is made on feature selection and feature extraction. e fast methods like SLFNs have been proposed in [4] which are based on extreme learning machine.
e optimization algorithms such as PSO [5] improve the efficiency of existing algorithms by targeting high values of fitness function with their respective mathematical model. e watermarking technique can be made more secure by adding encryption mechanism in it. e technique proposed in [6] adds security features to the abovementioned mechanism. e performance of the frequency domain techniques can be optimized using optimization algorithms like genetic algorithm [7]. e PSO algorithm [8] also optimizes the performance of watermarking technique by taking quality parameter into consideration. Graph Based Transform is a new kind of transform that interprets graph in the form of signal. A new transform based on graphs was proposed for depth map coding [9]. e process of frame selection is very important aspect of the research. e frame selection algorithm is proposed by using identical frame extraction concept [10]. e process of scene change detection was applied by Masoumi [11]. e proposed work is inspired from the research done over the years on the videos and intends to solve problems of existing research. Different frequency domain techniques are used in watermarking [12] but DWT has been used by many researchers. e application of SVD was to extract good number of features for performing transformation as largest coefficients in S component of SVD Matrix can resist image compression and processing attacks and embedding of watermark will not be affected when any of frequency domain methods is coupled with SVD [13]. Fourier Transform is also the part of Frequency Domain Technique [14] but the technique does not produce good results in terms of imperceptibility. e process of frame selection is done on the basis of number changes in scenes of the frames. e calculation is done using histogram difference [15]. A new Grey Wolf Optimizer is used to solve local optimum problems and find optimal solution from given set of solutions. GWO is an efficient PSO technique [16]. Hybrid combination of DWT-SVD was proposed by Security and Communication Networks various researchers. A hybrid technique based on DWT-SVD along with firefly algorithm got high values of quality parameters [17]. A video watermarking technique was proposed using multiple wavelets with the application of DWT-SVD [18]. GBT Transform is applied for data decorrelation [19] which is also an effective transform that can be applied to multimedia data. A semiblind DWT-SVD technique was proposed on compressed domain videos [20]. Graph Fourier Transform is used for depth map coding. is technique produces good results in multimedia data [21]. e hybrid transform DWT-SVD produces favorable results in terms of quality parameters. e hybrid transform is combined with Fuzzy BPN Architecture for grey scale images. It was producing good visual quality of watermarked image [22]. e DWT-SVD was applied on videos by Sharma [23]. e watermark embedding techniques hide the signal in the multimedia data but, to ensure security of data, various encryption techniques have been used along with frequency domain techniques. Wang [24] proposed encrypted watermarking technique using multiple kinds of chaos. A Hybrid Genetic Algorithm combined with fruit fly optimization [25] addresses QOS parameter that helps to solve the problem in less computation time.
e same technique can be used in watermarking to produce good results. A hybrid technique that combines BWT-SVD and optimization algorithm was proposed [26] to embed watermark in multimedia data. e blind H.264 compressed domain technique was proposed to find certain areas of the frames to embed watermark. Pattern recognition technique was proposed [27]. Sharma [28] enhanced the work by adding transpositional cipher in the combined transform of DWT-SVD to enhance security. Transpositional cipher used in [29] had issues in security. e cipher used in research [30] enhances security of any watermarking technique. Combined approach on Graph Based Transform and Singular Valued Decomposition was proposed for images in the respective work [31]. Cao [1] proposed an encryption technique that produces good results compared to others. e GBT-SVD Transform produces better results than GBT used in previous research [9,19]. Table 1 illustrates the gaps found in recent studies in this field.

The Proposed Methodology
In this section, we propose a frame selection mechanism followed by watermark embedding and application of certain attacks on the proposed technique. In this research, frame selection process is important as watermark information is sensitive that should not be leaked to any intruder. Embedding watermark in every frame makes the information easily accessible to unidentified user and adding watermark in every frame increases the size of watermarked video. erefore, frame selection mechanism is important. is mechanism is followed by watermark embedding and then evaluation of proposed technique is done by applying certain attack scenarios.

Frame Extraction and Selection.
e first phase in the proposed work is to find the suitable number of frames from extracted frames of the video. e process of finding suitable frames in real time is done using scene change detection. e watermark cannot be embedded on all frames of the video as it becomes every easy for any intruder to detect the watermark and add watermarking to all frames which also increases the size of the video. e process of finding suitable frames becomes significant. To select significant frames, scene change detection mechanism is applied. e comparison of adjacent frames with one another is performed. e grouping of identical frames is done. e value of the frame difference will decide whether frame will be considered as the part of the same group or different group. If difference is large, then it will be considered as part of different group. e parameter of decision will be taken as threshold; if the value of frame difference is higher than the value of threshold, the next frame will be the part of next group.
e same is illustrated in Figure 4. e temporal sampling is performed that enhances the process of frame selection that gives better results compared to [10]. e selection of the first frame is done from all different groups. Frame difference can be represented as histogram difference that can be expressed as where FD k is representing frame difference and T k is the histogram value of k th frame of level m and I is the number of levels of the histogram. e grouping of similar images is based on scene change detection. e threshold is maintained to detect intensity histogram difference to calculate sudden transition amongst frames (in order to find larger frame difference). is scenario is expressed as K b is threshold value. σ and µ are the standard deviation and mean value of selected frame intensity histogram difference. e selected value of α in the research is 2.8. e temporal sampling has also enhanced the process of frame selection. e criteria of frame selection depend upon the comparison of FD k with K b . e algorithm was tested on 6 videos. Relevant frame selection was done. e standard frame rate taken is 29.97. A total of 6 videos have been taken as data set for this process.
e videos with a greater number of scene changes will have a greater number of selected frames. is process is illustrated with the help of algorithm given as Algorithm 1.
e Akiyo video did not have any scene changes; hence, no frames will be selected and watermark embedding will not take place. Watermark embedding follows frame selection process only. e evaluation parameter of this step is total frame selection time from extracted frames of the video. e process of frame extraction is done followed by frame selection. Some videos have a smaller number of scene changes; hence, less frames will be selected. In case there is no detection of any scene done, then no selection of frame takes place. e pure storage video has higher number of scene changes; hence more frames will be selected. e importance of frame selection comes from the fact that watermarking on still number of frames will give a chance to any unauthorized person to get access of watermark content because of similar properties [32]. Figure 4 depicts the process of frames selection. Frame selection using scene change detection is giving better results especially in uncompressed domain. e grouping of similar images is done and threshold is calculated using (2); the moment scene change is detected, the first frame in the individual frame is selected and the same process follows till all the extracted frames are processed [33]. e process is fast avoiding similar frames to be selected, thus saving the time for frame selection and saving overall embedding time for embedding process. e results are formulated in MATLAB 2019b using i5 processor.

e Proposed Technique of Watermarking.
e next step after selection of frames is to embed encrypted watermark. Watermark is encrypted before it is embedded to selected frame [34]. Watermark embedding poses a great challenge of quality loss. To counter the problem of quality loss after embedding, the technique is supposed to be proposed that aims at high values of quality parameters like PSNR. Hybrid combination of Graph Based Transform, Singular Valued Decomposition, and Hyperchaotic Encryption is proposed to counter the security issues in multimedia data. Graph Based Transform (GBT) is a transform that uses signal in the form of graph and produces better results in terms of adapting signal structure of an image [35]. GBT is used as it is robust against various attack scenarios in the field of image processing. Singular Valued Decomposition counters the issues of dimensionality reduction [36]. After frame selection is done, selected frames are applied with GBT Transform followed by SVD and at the same time watermark is encrypted with Hyperchaotic Encryption before being applied to selected frame. e selected frame is taken as a signal in the form of a graph and transformation is applied using GBT [37]. e S value is taken after SVD is applied. e watermark is encrypted using Hyperchaotic Encryption and SVD is applied to it [38]. e S values of the selected frame and watermark are combined to form modified S value of watermarked frame. e proposed watermark embedding technique is further discussed in the following sections.

Embedding Technique.
Graph Based Transform is a newly formed transform that is represented by G � V; E; s { } where V and E are the vertices and edges of the graph, and s represents the frame signal [39] for graph G where m i, j represents the weight of the edge. e degree matrix D ∈ N × N is a diagonal matrix, where elements are en, the Laplacian-Graph Matrix L would be defined as where the operator L is also known as Kirchhoff operator, which is represented as adjacency matrix A. Eigenvalue decomposition is done to set of real nonnegative eigenvalues which are represented by Decorrelation of the signal defined on the graph is done using eigenvectors.

Reference
Main contribution Gaps Tabassum [10] In this research, identical frame extraction technique is proposed with 3-level DWT frame selection done using frame difference method. DWT is applied to higher band coefficients to get robustness against signal processing attacks. e quality is compromised and watermark embedding technique could be more efficient.
Masoumi [11] In this research, frame extraction is done by taking motioned part of the video; scene change detection is applied. Color separation of selected frame is done and watermark embedding is done in blue channel. Watermark is considered as pseudorandom numbers; each bit of watermark can be taken as scattered randomly through video frames in order to provide additional security feature.
e proposed algorithm becomes complex by applying a secured, encrypted technique.
Mishra [17] In this research, DWT-SVD technique is proposed along with optimization firefly algorithm using multiple scaling factors. e optimization adds to high values of quality parameters.
e technique is applied on grey scale images and additional security feature can be added.
Sridhar [18] In this research, hybrid DWT-SVD is applied on the videos with multiple wavelets. e efficiency of hybrid technique is always better than DWT.
e efficiency of frame selection algorithm is compromised.
Rajpal [29] In this research, fuzzy frame selection scheme with bidirectional extreme learning machine is done. Fuzzy rules are based on luminance, edge, and texture sensitivity. Fuzzy frame selection is based on scene change detection; weighting factor is based on these 3 parameters.
Security is compromised using transposition cipher.
Security and Communication Networks 5 Singular Valued Decomposition is done using equation number 10 where transform is done using S as it is more resistant to image processing attacks.

Encryption of Watermark before Embedding.
e watermark embedded on selected frames is encrypted using Hyperchaotic Encryption to add security feature to the proposed technique [1]. e value of x, y, z, and w calculated from above equation will be used for encrypting the watermark image to be used in a frame. e standard values of a, b, and c were taken as per the values in reference [1]. e second step is the conversion of R; S is done into x, y for column and row of the encrypted watermark image. e 3rd step is to interchange the coefficients of m th row and x(m) th row of image W − m � 1, 2, . . .i, N � 1, 2, . . .j; see Algorithm 2.
e encryption of a watermark image is represented as W(i, j) where image size is represented as m * n. e first step is generating the sequence of R, S using Lorenz system. e security feature added here adds to security feature by encrypting watermark before being embedded, thus making the technique more secure. Real time applications like broadcasting face security issues and copyright protection; the proposed technique combined with Hyperchaotic Encryption adds to security feature and also adds to copyright protection. Figure 5 depicts watermark embedding process.

Extraction Procedure.
e next section in the proposed work describes watermark extraction procedure so as to recover watermark from watermarked video. e extraction of a watermarked from watermarked video is a reverse process of embedding when watermark was embedded with the help of (13). e extraction of frames is followed by applying GBT and SVD and the extraction is calculated as per the following equation. is is followed by inverse GBT and inverse SVD; then decryption is done using a key; then watermark is recovered. Figure 6 depicts watermark extraction process: where W(i, j) is extracted watermark, WF′(i, j) is watermarked frame, and A(i, j) is selected frame. e extraction procedure is used to find the difference between original and extracted watermarks. High difference between both of the watermarks suggests that the technique is not efficient; however, as per result calculation, it was found that there is a negligible difference amongst both watermarks after extraction is done, as shown in Algorithm 3.

Performance Evaluation.
e performance evaluation of the watermarking technique is typically calculated in terms of quality parameters of the video and robustness against various attack scenarios such as Gaussian Noise, Sharpening, Rotation, Blurring, and JPEG Compression. e parameters are PSNR, SSIM, NC, and BER.
where G and H are rows and columns of the image: AI(i, j) is selected frame; EI(i, j) is watermarked frame.  Figure 4: Selected frames to be watermarked from different groups. selected frame. It is calculated using the following equation: (c) Structural Similarity Index Measure (SSIM): this parameter is used to find structural similarity between watermarked frame and selected frame. It is calculated from the following equation: where P m and P n represent average of m and n column; K m and K n represent variance of m and n; K mn represents covariance of m and n and c1 and c2 are variables.
(d) Bit Error Rate (BER): this is the inverse of PSNR calculated in the following equation:

Experimental Results
e results were evaluated in MATLAB 2019b using i5 processor. e frame selection time and embedding time are dependent on the type of processor used. e compiled results are dependent upon watermark embedding time and Input: Selected Frames Output: Watermarked Video Begin (1) for selected frames from 1 to k (2) Input binary image watermark (i, j).

(3)
Implement layer separation on selected RGB frame A(i, j) and watermarked image W(i, j). (4) Encrypt the watermark image W(i, j). (5) Convert selected frame into GBT Transform on each layer of A(i, j). (6) Use SVD Transform and extract S Feature of USV of A(i, j) and W(i, j).

(7)
Target S value from both sides to be merged. (8) Embed watermark W(i, j) to A(i, j) using value α as 0.02 using equation Get modified S value from both A(i, j) and W(i, j) (10) Repeat steps from 1 to 9 till all the selected frames are processed. (11) End for (12) for frames 1 : m (13) Replace watermarked frames with selected frames from the directory. (14) Use same frame rate to combine all frames including watermarked and rest of extracted frames (15)   frame selection time. A total of 6 Common Interchange Format (Cif ) encoded videos have been taken and frame selection mechanism entirely depends upon number of scene changes in the video. Some videos have a greater number of scene changes; hence more frames will be selected. Akiyo did not have sufficient scene change detection so the watermarking technique could not be applied on that as the value of FD k (frame difference) was not greater than K b (threshold) so no significant frames were selected from the video; rest of the videos have significant frames selected as per frame selection algorithm.. e data sets of the videos were obtained from Figures 7(a)-7(e) which signifies some selected frames from the data set of videos. Along with these videos 2 binary watermarks and their encrypted versions have been shown; the compressed domain videos taken in the research are the same type of videos used in broadcast application; to remove unauthorized access to these videos, the given videos are embedded with encrypted watermark that addresses the issues faced by real time application. e encrypted watermark not only addresses security issues but also adds to copyright protection to achieve ownership identification.

Input: Watermarked Video Output-Recovered Watermark Begin
(1) Input watermarked video (2) Apply frame extraction mechanism using frame rate 29.97 (3) for frames from 1 to k (4) Implement layer separation of watermarked frame (5) Convert Watermarked frame into GBT Transform on each layer of WF′(i, j). (6) Perform extraction of USV feature of each layer of WF′(i, j).
Extract S value of watermarked frame.
Extract S value of the watermark (10) Perform inverse SVD to combine S value with USV of each layer. (11) Perform inverse GBT Transform of watermark images (12) Apply decryption mechanism to get extracted watermark (13) Repeat Steps 4-12 to get extracted watermark.

Security and Communication Networks
Format and it is referred to as a standardized format for picture resolution and the data has been obtained from website named https://media.xiph.org/video/derf/. e value of quality parameters is taken as per comparison with original and watermarked frames. Table 2 represents the comparison of the input videos and the number of frames selected from the given videos. It was found that Pure Storage video has higher number of frames selected out of all videos. Table 3 represents the embedding of watermark 1 on selected frames without any attack. e performance of the proposed technique is calculated with various factors represented in Table 3. Figures 9(a)-9(d) describe the performance of embedding technique against no attacks applied to it. Table 4 compiles the processing time (in seconds) required to carry out frame selection, embedding time taken for the given set of videos. e time is entirely based on processor requirements. e total time consumed depends upon selection of frames from the video. Pure Storage video has got 5 frames selected and the time for every frame varies from 20 to 35 seconds for every frame. e value of embedding time is directly proportional to number of selected frames. Total of 5 frames were selected from Pure Storage video; thus, total embedding time is the highest for the same video. e watermark embedding factor is kept being 0.02 and GBT was followed by SVD on selected frames and mixed with S value of watermark. e proposed technique is fast and, as per processor requirements, works considerably at good speed. e plots in Figures 10(a) and 10(b) signify time taken for selection of frame from 5 videos. More number of changes in the video is directly proportional to the frame selection time and watermark embedding in selected frames for a single video is dependent upon number of frames selected. e plot in Figures 10(a) and 10(b) signifies the embedding time taken by selected frames from the video. Table 4 represents the total frame selection time and embedding time of the input videos.

Processing Attacks.
e robustness of the proposed technique is tested against various attack scenarios such as Gaussian Noise, Sharpening, Rotation, Blurring, and JPEG Compression. A series of experiments have been conducted to attack every watermarked frame to measure quality loss. e robustness of the technique entirely depends upon the values of PSNR, SSIM, NC, and BER.       to low and high frequency bands of the image. In our research, this attack is applied to find out difference in watermarked frames with this attack and without it. e results of quality parameters after Sharpening Attack are represented in Table 6.

Rotation Attack.
In Rotation attack, a watermarked frame is rotated with an angle of 90 using watermark 1. Higher value of Rotation attack will affect PSNR of the watermarked frame. e quality metrics of Rotation attack is affected by the higher angle in which the frame is rotated. It can be seen from plots in Figures 13(a)-13(d) that average PSNR, NC, and SSIM deteriorate with increase in attack value and BER increases with increase in attack value. e Rotation attack is carried out by rotating the watermarked frame and normal selected frame. e technique is vulnerable against Rotation attack as it does not achieve good results in that attack. e addition of optimization algorithm   to find best fitness function can improve the values of quality metrics against this attack. Table 7 represents results of quality parameters after applying Rotation attack.

Blurring Attack.
In Blurring attack, a random sequence of real values {2.05} is added to all frames of the watermarked video using watermark 1. e Blurring attack is caused by motion of an object. e more the object is moved, the lower the value of PSNR will be. It can be seen from plots in Figures 14(a)-14(d) that average PSNR, NC, and SSIM decrease with increase in attack value and BER increases with increase in attack value. In the research, we applied Blurring attack to check the motion of watermarked frame. Higher values of PSNR will indicate effectiveness of the technique. Table 8 represents results of quality parameters after applying Blurring attack.

JPEG Compression Attack.
In JPEG Compression attack, value {98} is taken and applied to all frames of watermarked video. JPEG Compression number decides how much compression attacks can be applied. JPEG Compression application on watermarked frame indicates no significant change. It can be seen from plots in Figures 15(a)-15(d) that average PSNR, NC, and SSIM decrease with decrease in value of compression attack value and BER increases with decrease in attack value. Table 9 represents results of quality parameters after applying JPEG Compression attack.

Conclusion and Future Work
We proposed a novel frame selection based watermarking technique (GBT-SVD-hyperchaotic) to address quality loss of data. Frame selection algorithm is proposed to select appropriate number of frames as addition of watermark in every frame leads to time complexity of the embedding algorithm. Frame selection is done on the basis of number of scene changes done in the video. e hybrid combination of Graph Based Transform, Singular Valued Decomposition, and Hyperchaotic Encryption provides efficient results for watermark embedding. e proposed technique was found to be robust against many signal processing attacks, Gaussian Noise, Sharpening Attack, Rotation, Blurring, and JPEG Compression. e additional security mechanism applied in the proposed work gives added advantage over transpositional ciphers in related work. e proposed technique is fast; however, it faces the limitation of absence of optimized algorithms. e performance of the proposed technique can be improved by applying optimization algorithms like Grey Wolf Optimization that will optimize the embedding factor, thus targeting high values of PSNR.

Data Availability
e data are open and available on request.

Conflicts of Interest
e authors declare that there are no conflicts of interest.