Optimization based interesting region identification for video watermarking

Journal of Information Security and Applications 49 (2019) 102393

Contents lists available at ScienceDirect

Journal of Information Security and Applications journal homepage: www.elsevier.com/locate/jisa

Optimization based interesting region identification for video watermarking Amir M. Usman Wagdarikar∗, Ranjan K. Senapati Dept. of ECE, K L E F (Deemed-to-be-University), Vijayawada, Guntur, Andhra Pradesh, India

a r t i c l e

i n f o

Keywords: Chronological concept Digital video watermarking Embedding MS algorithm Multimedia

a b s t r a c t The advancements in multimedia watermarking technologies have gained lots of attention in the past few years. A digital video watermarking is the data embedded in the video and sent to the recipient. Privacy is considered as the major issue in digital media while hiding the data by maintaining the video quality. This paper presents an optimization algorithm for video watermarking based on interesting regions. Here, the optimal regions for the video watermarking are selected using the proposed ChronologicalMoth Search (Chronological-MS) approach, which is established by modifying the Moth Search (MS) algorithm using chronological concept. The employed fitness function contains cost function, which uses several parameters, like energy, edge, pixel intensity, brightness, and coverage. At first, the input video file is subjected to feature extraction, and the extracted features are employed for selecting the optimal regions using the proposed Chronological-MS. After that, the wavelet transform is given to the original image to obtain the wavelet coefficients. Here, the privacy message is divided into binary images using bit plane technique. Thus, the embedding process is carried out for hiding the secret message with the use of interesting regions identified using the proposed Chronological-MS algorithm and is retrieved at the extraction phase. The performance of the proposed Chronological-MS shows superior performance with correlation coefficient and PSNR of 1.00 and 98.741 dB, respectively. © 2019 Elsevier Ltd. All rights reserved.

1. Introduction Video watermarking is the attractive domain for developing a system with authentication and copyright protection method. Here, the security is considered as a major problem, which includes copyright violation of multimedia, tampering, and illegitimate fraud [1]. The cryptographic mechanisms are employed for transferring the data securely between the content provider and customer [2]. The multimedia applications have increased the demands for designing secure mechanisms for legally distributing digital data. The transmission of multimedia data becomes simpler due to high internet speed and multimedia systems in a distributed environment. Thus, there is a need for securing the copyrights of digital content [3]. Watermark is adapted in original multimedia data for refraining the originality of data, and the owner can interpret the data. The watermarking method is fed to audio, video [35,37], and multimedia applications. The techniques may hinder the originality of the data due to the rapid increase in internet technologies. The distribution, duplication, and the access of multimedia data are simpler, which results in multiple issues,

∗

Corresponding author. E-mail address: [email protected] (A.M.U. Wagdarikar).

https://doi.org/10.1016/j.jisa.2019.102393 2214-2126/© 2019 Elsevier Ltd. All rights reserved.

like broadcasting and illegal usage. Thus, the safety of multimedia content has become a major challenge [4]. Due to huge data and intrinsic redundancies between video signals and frames, there is a chance of extremely vulnerable privacy attacks that include statistical analysis, frame averaging, frame swapping, and frame dropping. The existing algorithms were unable to address the problems for securing the multimedia contents [13]. The watermark represents a digital code implanted in the video that can be utilized for the embedded broadcast of binary information and depicts the copyright owner. Watermarking is utilized for representing the identification of a legal receiver for each copy [14]. The protection of copyright is considered as the most important application of watermarking approaches [12]. Watermarks are usually embedded into pixel and convert coefficient domain. In pixel domain techniques, the host image pixels are adjusted concerning watermark bits and are simpler to execute with huge payload and low computation cost. But, most of the pixel domain mechanisms provide less robustness. In coefficient domain techniques, the coefficients frequency of any transform is adjusted with respect to the watermark bits. Some watermarking techniques use the image segmentation [34,46,47] and optimization algorithms [36]. Various coefficient domain watermarking approaches provide improved robustness with reduced payload and increased computational complexity [15].

2

A.M.U. Wagdarikar and R.K. Senapati / Journal of Information Security and Applications 49 (2019) 102393

For fulfilling multipurpose applications, various watermarking algorithms [10] that are based on Fourier transform [7], and wavelet transforms [6], are employed. The multipurpose watermarking includes a cocktail or multiple watermarking methods [8], which are applicable for copyright protection by embedding different watermarks and are robust to different attacks. Some of the robust watermarking technique depends on Vector Quantization (VQ) [9], which inserts the watermark data to the encoded indices under several constraints that indicate extra distortion lesser than a given threshold [11]. Generally, watermarking embedding and the retrieval should contain low complexity as real time watermarking is enviable [12]. Depending on the domain in which the watermark is embedded, watermarking methods are classified into three types, namely compressed, spatial, and transform domain watermarking. In compressed domain watermarking methods, the watermark is embedded in an encoded bit stream generated using encoders. The spatial domain watermarking is achieved by directly modifying the pixel values of the embedding channel of a video frame. In a transform domain watermarking system, the host frame in a video sequence is converted to a new domain before embedding the watermark. This conversion is obtained by the most commonly used transforms, like dual-tree complex wavelet transform (DT CWT) [38,39], discrete wavelet transform (DWT) [40,41], DCT [42,43], and singular value decomposition (SVD) [44,45]. At the watermark embedding process, the transform domain coefficients are modified by the watermark and then an inverse transform applied on these modified coefficients to generate a watermarked frame. These types of techniques are vigorous, secure, and provide more imperceptibility than spatial domain-based techniques [32]. This research paper presents a transform domain video watermarking technique by proposing an optimization algorithm, named Chronological-MS, for selecting the optimal regions to be embedded in the video. The fitness function of the proposed Chronological-MS algorithm uses a cost function that contains several parameters, namely edge, energy, brightness, coverage, and pixel intensity. Here, the proposed Chronological-MS is developed by incorporating Chronological concept in Moth Search (MS) algorithm and employs wavelet transform for embedding and extraction of secret message. The input video consists of several frames where the secret message is embedded based on the regions selected using the proposed algorithm. At first, the frames extracted from the input are applied to the wavelet transform for obtaining the wavelet coefficients. Here, the bit plane method is applied to a secret message for generating the binary images. At last, the extraction process is carried out at the receiver side for extracting the secret message with the watermarked image. The main contribution of the paper is devising an optimization algorithm, Chronological-MS, by altering the update process of MS algorithm using an idea, named chronological, for determining the suitable regions for embedding the message in the video frame. The rest of the paper is structured as follows: Section 1 depicts the introduction of watermarking and multimedia application. Section 2 explains the literature survey considering eight research works using watermarking and the challenges faced by the techniques. Section 3 describes the proposed ChronologicalMS for video watermarking. Section 4 explains the results of the Chronological-MS, comparing its performance with various previous works, and finally, Section 5 deliberates the conclusion. 2. Literature review This section describes the literature review of eight different methods utilized for performing video watermarking along with their challenges. Alper Koz et al. [19] designed a technique, named spread spectrum technique for the Human Visual System (HVS) based on video

watermarking. This technique makes use of a temporal dimension through the temporal sensitivity of HVS. The temporal contrast and the threshold value are described for enhancing the watermark. The spread spectrum mechanism attained improved robustness with respect to noisy pixels, temporal shifts, and frame rate conversions, but the method faces certain complications while employing the HVS systems. Komwit Surachat et al. [20] developed a pixel-wise digital video watermarking technique utilizing a Weiner filter. Here, the embedding is done in chrominance channel for the video frame. During extraction, the filter uses a 3 × 3 window size for improving the quality of watermarking. The result generated by the pixel-wise digital video watermarking technique offered better performance and thus, improves the robustness from security attacks, but the extraction influences the accuracy. Samira Mabtoul et al. [21] developed a Singular Value Decomposition (SVD) using a watermarking algorithm based on complicated wavelet transforms. The method considers the input data as a color image that comprises YCbCr color components. At first, the color component of every video frame applies 2-level decomposition of Dual Tree-Complex Wavelet Transform (DT-CWT) transform to obtain the subbands and then, the SVD is finally applied. The obtained embedded image is robust against blur, histogram equalization, scaling, and Gaussian noise. In several cases, the SVD based techniques alleviate the rate of embedding. Hui-Yu Huang et al. [22] designed a pseudo-Three Dimensional Discrete Cosine Transform (3D DCT) and quantization index modulation to initiate video watermarking. Here, the input frames were chosen on the basis of blocks, in which the message was embedded. The pseudo-3-D DCT used DCT transformations for evaluating the factor and for recovering the hidden messages efficiently. Accordingly, the data is entrenched in the quantization regions of the frame using Quantization Index Modulation (QIM), but the method was susceptible to several attacks, like geometric attacks that include scaling and rotation. Sake and Tirumala [3] developed a method by employing Biorthogonal Wavelet Transform (BWT) and SVD for protecting the copyrights of images. Two main processes, which include watermark extraction and watermark embedding processes, are employed for improving the efficiency of video watermarking. After embedding, the input video sequences are transformed to a total number of frames. Artificial Bee Colony (ABC) approach is adapted in BWT to produce random frames for initiating the embedding process in watermark video sequences. Further, the extraction of the watermarked image, which is considered as the reverse process of the watermark embedding, is performed where the watermark image is extracted from the video sequences. Shukla and Sharma [16] designed a scene based video watermarking technique by adapting discrete wavelet transform for protecting the video copyrights. The technique integrates video watermarking with Successive Estimation of Statistical Measure (SESAME) technique. For reducing the computation complexity, the watermark is embedded in the scene change frames. This technique focused on correlation-based scene change detection method. However, the method failed to consider a secure method to provide copyright protection for the embedded video. Naseem.et al. [17] developed a block-based transform domain technique based on Fuzzy Rule Based System (FRBS), which chooses an image from a test image to embed and hold the desired capacity using high robustness. FRBS contains two phases, in which the initial process selects candidate image blocks, and the second process selects the coefficients from the chosen candidate blocks to embed the desired capacity. At last, the image is chosen as a candidate image, which contains improved Peak Signal to Noise Ratio (PSNR) and correlation values with equal desired capacity. The technique failed to increase the capacity of data embed while

A.M.U. Wagdarikar and R.K. Senapati / Journal of Information Security and Applications 49 (2019) 102393

retaining high imperceptibility and robustness and suffers from high computational complexity. Thongkor et al. [18] designed a digital image watermarking with the images captured by the camera. In this method, each component of the image pixels is utilized to embed a user watermark bit for providing largely embedded watermark. Once the watermark is printed, the extraction process is carried out for reducing the distortions from the watermark image component. The robustness of the image is demonstrated considering various types of attack distortions. 2.1. Challenges The challenges faced by existing techniques are elaborated as follows, •

•

•

•

The video watermarking methods face several challenges, like a huge amount of redundant data among frames, the imbalance between motionless and motion areas, video broadcasting, etc. Moreover, video watermarking techniques are more vulnerable to pirate attacks that include statistical analysis, frame averaging, frame swapping, and lossy compressions [5]. Digital watermarking suffers from various problems, which include content authentication and copy protection. The techniques utilized for digital watermarking are designed for a single purpose, which is a major challenge [11]. The video watermarking techniques are more complex because of its huge volume and innately redundant data between the video frames. The existing video watermarking techniques do not offer robustness with respect to the geometrical attacks, which involve sharpening, scaling, blurring, and compression [16]. The digital images are employed in digital watermarking. The difference between illegal copies and legal copies is difficult, and it directly defects media creators and dispirits them from designing advanced applications [18].

3. Proposed chronological-MS and wavelet based video watermarking algorithm The major goal of this research is to hide the message to a multimedia source as video file without affecting the superiority of the video file. It was performed with embedding a watermark into the video frame so that intruders could not access the confidential information within the video file. The aim of the watermarking technique is to make the watermarked file more robust and secure. Fig. 1 illustrates the block diagram of video watermarking using proposed Chronological-MS. For initiating the embedding process, the interesting regions are determined using the proposed Chronological-MS algorithm, which is developed by modifying the MS algorithm, based on the chronological concept. The technique uses a wavelet transform for extracting and embedding the secret message from the video frames. Here, the bit plane approach is utilized for generating the binary image for embedding and extraction. After determining the optimal region in the video frame, the embedding of the message is done. At last, for extraction, the receiver retrieves the concealed data using the fitness value.

3

3.1.1. Solution representation Solution encoding plays a significant role in determining the optimal solution in optimization problems. It helps the algorithms to attain the optimal solution from a set of solutions. The solution of the proposed Chronological-MS represents several regions, from which the best regions to be embedded are chosen using the fitness function. The solution is encoded as a vector, wherein |E| denotes the total regions in the frame, from which the optimal e regions are selected by the proposed algorithm. The cost function developed in [26] is utilized as the fitness function in the proposed Chronological-MS algorithm to find the optimal regions. 3.1.2. Algorithm The proposed Chronological-MS is developed by integrating the chronological concept into the MS algorithm for selecting the optimal regions. MS algorithm [25] is an advanced metaheuristic algorithm motivated with photo axis and levy flights of the moths. MS algorithm can search for the best solution effectively with improved accuracy. Moreover, the algorithm negotiates complex operations, and thus, the execution of the MS algorithm is easy and flexible. The chronological concept is adapted to update the solution based on the past events that had occurred with respect to the time. Hence, the incorporation of chronological concept into the MS algorithm is made for finding the optimal regions. The steps involved in the proposed Chronological-MS are described as follows, 3.1.2.1. Initialization. Initially, the population U is initialized in a random manner, the generation number is fixed asu = 1. Also, the acceleration factor is defined asϕ , maximum generation is defined asw, and max walk step is defined as θ . The initialization of the population is represented as,

U = {U1 , U2 , . . . , Uu , . . . , Uw }

(1)

where, w is the population size; where 1 ≤ u ≤ w. 3.1.2.2. Evaluation of fitness function. The fitness function is computed for the individual solution using the Eq. (11) for better result. The output having improved fitness value is considered as the optimal output. The best solution is determined at the previous iteration as each solution craves to obtain a better position. 3.1.2.3. Solution updation. According to the MS algorithm, moths with lesser distance from the better one flies around optimal one in the form of Levy flights. Moreover, levy flights performed to update the position. Hence, the update solution cording to MS algorithm is given as,

the the are ac-

v − U v )) Uuv+1 = β × (Uuv + φ × (Ubest u

(2)

where, β denotes scale factor, Uuv denote the position of a moth at v denote the best generation u and ϕ denote acceleration factor, Ubest position of a moth at generation u. The solution for the previous iteration is given by, v−1 − U v−1 )) Uuv = β × (Uuv−1 + φ × (Ubest u

(3)

After substituting Eq. (3) in Eq. (2) v−1 − U v−1 ) + φ × (U v − U v )) Uuv+1 = β × (β × (Uuv−1 + φ × (Ubest u u best

(4) 3.1. Proposed chronological-MS algorithm The problem of deciding the regions in the frame to be extracted is considered as a search problem. Here, the optimal region is determined using an optimization based searching algorithm, named Chronological-MS, which is developed by the integration of chronological concept into the MS algorithm [25].

After rearranging the above equation, v−1 − U v−1 ) + βφ × (U v − U v )) (5) Uuv+1 = β 2 × (Uuv−1 + φ × (Ubest u u best

The solution for (v + 1)th iteration is given based on the chronological concept as,

Uuv+1 =

Uuv+1 + Uuv+1 2

(6)

4

A.M.U. Wagdarikar and R.K. Senapati / Journal of Information Security and Applications 49 (2019) 102393

Fig. 1. Block diagram of video watermarking using proposed Chronological-MS algorithm.

The final equation obtained after substituting Eq. (2) and Eq. (5) in the above equation is given by,

Uuv+1 =

1 v − U v ) + β 2 × (U v−1 + φ β × (Uuv + φ × (Ubest u u 2  v −1 v −1 v v × (Ubest − Uu ) + βφ (Ubest − Uu

(7)

The update position is obtained in the MS algorithm for finding the most appropriate location of the search agent. Therefore, the final solution update equation is expressed by,

Uuv+1 =

β

v − U v ) + β × (U v−1 + φ {(Uuv + φ × (Ubest u u 2 v−1 − U v−1 ) + φ × (U v − U v )} × (Ubest u u best

(8)

3.1.2.4. Finding a feasible solution. The updated solution is determined based on the fitness metric in Eq. (2), for checking the feasibility of the original solution. The fitness function of the next solution is combined, whether that of the previous solution is optimal. Therefore, the location is updated until the better result is obtained. 3.1.2.5. Termination. The process is iterated repeatedly till vmax for finding the best position. 3.2. Video watermarking using the proposed chronological-MS algorithm This sub-section presents the video watermarking using the proposed chronological-MS algorithm. The video frames consist of a total number of frames where the watermark needs to be embedded. The developed technique contains four following steps: i) Extraction of keyframes, ii) Computation of fitness function, iii) embedding process, and iv) Extraction process. At first, the keyframes are removed from the video image file with equal intervals of time, and then, the wavelet transform is subjected to each of the frames.

3.2.1. Extracting frames from the video file The watermark is hidden into the regions of the video frame using a newly designed optimization algorithm. Let the input database be denoted as I, which containy videos, and is represented asI = {Wd ; 1 ≤ d ≤ y}. Each video containsf frames, and is expressed asW = {Gh ; 1 ≤ h ≤ f }. Let the message be indicated asPthat contains m rows and lcolumns and is formulated asP = hm,l . From the f frames, the most suitable keyframe is employed for initiating two processes. The frames are extracted at equal intervals of time that is represented as f/h, where hsignifies the integer value. Hence, the keyframe is removed from the video file is formulated as,



W = Sr ;



1≤r≤s

(9)

where, Srefers to the original frame utilized to embed the secret message.

3.2.2. Bit plane slicing for partitioning the messages Assume Pbe the secret message, which is transferred to the target receiver over the network. For enhancing the security, the message is embedded in the video files. Generally, the frame consists of gray values from 0 to 255 represented in 8-bit value. In bit plane slicing [24], the stack of binary images is taken into consideration. Here, the images at the bottom are considered as the least significant, and the images at the top are considered as the most significant. The bit plane technique is to partition the raw frame image to eight-bit planes where the plane 0 is the least significant bit, and plane seven is the most significant bit. The bit plane technique provides eight binary images of the private message, which is represented as,

P = {P1 , P2 , P3 , . . . , P8 }

(10)

where, P1 indicates the first binary image and P8 denotes the 8th binary image of the secret message.

A.M.U. Wagdarikar and R.K. Senapati / Journal of Information Security and Applications 49 (2019) 102393

3.2.3. Devising fitness function to find the optimal region Here, the fitness function is computed using the cost function, which includes several parameters, like brightness, edge, wavelet energy, pixel intensity, and coverage. The cost function is utilized for providing the position, in which the secret data is to be embedded. The fitness function is given as follows, |E | X 1  N= Da,b |E | a

(11)

5

the maximum grayscale value. Thus, the cost value of the contrastenhanced image is represented as, |E |    1 StT (a, b) − StT( p) (a, b) 255 ∗ |E |

BJ (a, b) =

(17)

p=1

where, StT (a, b) refers to the current pixel and StT( p) (a, b) refers to the neighborhood pixel in a region.

b

where, |E| indicates the number of regions in the grid, andX represents the grid size. Da,b denotes the cost function [26] that is formulated as,

D=

  ( 1 − S z ) + ( 1 − Az ) + 1 − BJ + Bx + C k

(12)

4

where, Sz ,Az ,BJ ,Bx , and Ck , represent the edge, the pixel intensity, the wavelet energy, the brightness, and the coverage corresponding to the grids. 3.2.3.1. Pixel intensity. Here, the pixel values of each grid in the video frames that serve as an accurate depiction of original frames are taken into consideration. Each pixel represents an intensity value, and thus, the pixel intensity is considered as an important aspect for evaluating the fitness. Hence, the pixel intensity is formulated as given below,

Stz (a, b) =

1 255 ∗ |E |

|E |  

 St (a, b) − Stp (a, b)

(13)

p=1

p

where, St (a,b)denotes the value of the current pixel, St (a, b) indicates pth neighborhood pixel, and p represents the number of neighborhood pixels. 3.2.3.2. Edge detection. It is the process of determining points in a video frame, in which the brightness of image varies instantly such that the pixel deviates from low intensities to high intensities and vice versa with certain discontinuities. The purpose of edge detection is to determine and capture essential events and to change the image property. The edge is computed by adapting the canny edge detector [23]. The canny edge detection algorithm employs four steps, which include i) Non-maximum suppression, ii) Gradient searching iii) Smoothing iv) Edge linking and v) double thresholding. The edge detector is formulated as

A = edge(St )

(14)

where,Srepresents the input. Thus, the edge with cost value is represented as follows,

Az (a, b) =

|E | 1  A p (a, b) |E |

(15)

p=1

where, A p (a, b) defines the edge pixels in each grid or region. 3.2.3.3. Brightness. The aim of image enhancement is for improving the perception of information for human viewers and to offer the best input for other image processing mechanisms. The enhancement of images is done using noise reduction, sharpening edges, filtering, and contrast manipulation. The contrast enhancement [24] depends on pixel value using local maxima and local minima. Thus, the contrast enhancement, which indicates the brightness, of the video frame is represented as,

StT (a, b) =

St (a, b) − M1 × Me M2 − M1

(16)

where, StL (a, b) denote the contrast-enhanced frame image, M1 represents the minima and M2 indicates the maxima and Me represents

3.2.3.4. Wavelet energy. For evaluation, the conventional wavelet transform [23] is employed for extracting the wavelet coefficients, as it exhibits the scale features and positions. The wavelet transform is utilized for decomposing the raw frame to four sub-bands, which includes HL, HH, LL, and, LH represented as follows,

x = DW T (S )

(18)

Applying DWT, the frame image is divided into vertical, horizontal, diagonal, and. Approximation. Here, the LL band is approximation band, which is similar to the original frame. The second level decomposition is carried out for obtaining the LL band. After that, the LL band is divided to b blocks, for which the energy is computed. Therefore, the block energy of wavelet coefficients is represented as follows,

xJ ( j ) = −

T 

[qa log (qa ) + 1]

(19)

a=1

where, Tdenotes the total pixels in a block, and qa indicates the probability measure.



Bx (a, b) =



1 α zJ ( j ) ;

j ∈ (a, b)

(20)

where, α represents the normalization factor varies from 0 to 1. 3.2.3.5. Coverage. The coverage is the ratio of same neighborhood pixels into the number of pixels considering the image frame W. The cost of coverage is computed for entire wavelet coefficients. It is formulated as: p

C k (a, b) =

N St (a,b) W

(21)

p

where, N St (a,b) represents the number of p similar neighborhood pixels, and W denotes the number of pixels in the frame image. Thus, the optimal region, selected using the proposed Chronological-MS algorithm, is utilized for embedding the secret message. 3.2.4. Wavelet transform based embedding After the computation of cost values for each region, the optimal region is selected for initiating the watermarking process. The embedding is utilized for hiding the secret message into the video frames. Then, the bit plane technique is employed for acquiring eight binary images, as the wavelet produces eight subbands. Thus, the wavelet transform is applied for extraction and embedding process. During embedding, the input frame is divided into eight sub-bands by adapting two level decompositions, and the watermark is embedded to wavelet coefficient. 3.2.4.1. Binary fitness value. The fitness function is utilized for evaluating the fitness value using different criteria. The computation provides the location in which the message needs to be embedded. Here, the watermarking is employed for evaluating the binary cost value to extract and embed the private message. Thus, the message bit of each bit plane image is embedded in every wavelet coefficient.

6

A.M.U. Wagdarikar and R.K. Senapati / Journal of Information Security and Applications 49 (2019) 102393

3.2.4.2. Embedding. The input video is utilized for embedding the data. The YUV spacing approach is applied for converting the image from the color image as the video frame is a color image. Thus, the representation of a YUV image depicts Y as a grayscale image and two chrominance components (V-red projection and U-blue projection). Thus, the Y component is given as the input. 3.2.4.3. Wavelet transform. The wavelet transform is utilized for acquiring the wavelet coefficient in both phases. Here, the one level decomposition is subjected to attain HH, LH, LL, and HL. The wavelet transform has several advantages, such as data structure similarity considering resolution and decomposition levels. The DWT is applied as a multistage transformation. Here, the image is partitioned into four subbands using level 1 in the DWT domain. The LL denotes the coarse-level coefficients, and HL, HH, and LH depict the finest scale wavelet coefficients. Here, the LL subband can be further processed for obtaining another decomposition level. Moreover, the watermark is more likely to be embedded in LH, HL, and HH subbands for maintaining the image quality. The subbands are represented as,

{LL, LH, HL, HH} = DW T (S )

Initially, the first level decomposition of the wavelet transform is given to the video frame to attain HH, LL, HL and, LH, which is represented as,





LL , LH  , H L , H H  = DW T (ω∗ )

(27)

ω∗ denotes

where, the embedded image. Further, the second level decomposition is employed on every subband and is represented as,

{LL∗1 , LH1∗ , HL∗1 , H H1∗} = DW T (LL∗ ) {LL∗2 , LH2∗ , HL∗2 , H H2∗} = DW T (LH ∗ ) {LL∗3 , LH3∗ , HL∗3 , H H3∗} = DW T (H L∗ ) {LL∗4 , LH4∗ , HL∗4 , H H4∗} = DW T (H H ∗ )

(28)

At last, the embedded image consists of wavelet subbands and is denoted as E ∗ = {LL∗1 , H H1∗ , ........, LL∗4 , H H4∗} and is employed for extracting the hidden message. The extraction process is formulated as,

Q (a, b) = Ek∗ (a, b) − Ek (a, b)

(29)

where, Q denotes the retrieved message.

(22)

where,Sdenotes frame and the wavelet coefficients of various subbands are denoted as {LL, LH, HL, HH}. Further, the second level of decomposition is fed to each band to obtain 16 subbands, represented as,

{LL1 , LH1 , H L1 , H H1} = DW T (LL ) {LL2 , LH2 , H L2 , H H2} = DW T (LH ) {LL3 , LH3 , H L3 , H H3} = DW T (HL ) {LL4 , LH4 , H L4 , H H4} = DW T (H H )

(23)

The HH and LL wavelet coefficients are used in the wrap image to initiate the extraction and embedding process. The embedding is performed on the optimal regions obtained from the proposed Chronological-MS. The embedding function is determined using the formula,

Fk (a, b) = Ek (a, b) + g ∗ Pl (a, b)

(24)

where, k and l represents the message bit plane and wavelet bands, where 1 ≤ k ≤ 8 and 1 ≤ l ≤ 8. Fk (a, b) denotes the watermarked image, Ek (a, b) indicates the cover image, Pl (a,b) denotes the lth bit plane message and gdenotes the embedding strength. Inverse Wavelet Transform: After acquiring the embedded image, it is given to inverse wavelet transform for obtaining the hidden secret message. The wavelet based embedded image is represented asF (a.b) = {LL1 , H H1 , ........, LL4 , H H4 }. Then, the decomposition of the inverse wavelet transform is expressed by,



IDW T (F (a, b) ) ⇒ ω (a.b) = LL , LH  , H L , H H 



4. Results and discussion The results acquired by the developed Chronological-MS using the wavelet transform based video watermarking are described in this section. The proposed Chronological-MS using wavelet transform is analyzed based on the performance evaluation measures and is compared with the existing techniques. 4.1. Experimental setup The proposed Chronological-MS using wavelet transform is executed in the MATLAB framework. The system uses Windows 10 OS with 4GB RAM and i5 processor. 4.1.1. Dataset description The experimentation is done using three videos. Video 1 and video 2 are extracted from the online source, YouTube. Video 3 is taken from the public dataset UCF-Crime [33]. The input taken into consideration for initiating video watermarking is the sample of videos taken from YouTube. The two videos are given as input, where the secret message is embedded in a secure manner. 4.1.2. Evaluation parameters The performance of the developed Chronological-MS using wavelet transform is analyzed based on Peak-to-Signal Noise Ratio (PSNR) and correlation coefficient. The metrics are formulated as follows,

(25)

where, ω(a.b) denotes inverse wavelet transform of the embedded image. Then, the second level of decomposition of the image ω(a.b) is expressed by,

4.1.2.1. PSNR. The PSNR is computed for defining the superiority of the embedded image. The PSNR is formulated as the ratio of the original image into the watermarked image.

IDW T (ω (a, b) ) ⇒ ω∗ (a.b)

4.1.2.2. Correlation coefficient. The correlation coefficient offers statistical relationship among the original image and embedded video image.

(26)

where, ω∗ (a.b) is the embedded or watermarked image. 3.2.5. Secret message retrieval Once the embedded image is determined, then the image is transmitted through a communication channel for reaching the target. The fundamentals required for the extraction process contain the original image, watermarked image, and fitness function. The wavelet transform is applied for obtaining wavelet coefficients. At the extraction steps, the receiver extracts the secret message.

4.2. Comparative methods The comparative analysis is achieved by analyzing the performances of the comparative techniques, Wavelet method [28], Least Significant Bit (LSB) [27], LSB [27] +Cost, Wavelet+Cost [26], with that of the proposed Wavelet + Chronological-MS using the two metrics.

A.M.U. Wagdarikar and R.K. Senapati / Journal of Information Security and Applications 49 (2019) 102393

7

Fig. 2. Experimental results for video 1 (a) Original frames of sample input videos (b) Secret image (c) Embedded image (d) Salt and pepper noise added image e) Extracted image.

8

A.M.U. Wagdarikar and R.K. Senapati / Journal of Information Security and Applications 49 (2019) 102393

Fig. 3. Experimental results for video 3 (a) Original frame of sample input video (b) Embedded image (c) salt and pepper noise added image (d) Extracted secret image.

Fig. 4. Experimental results for video 3 using temporal attack (a) Original frame of sample input video (b) Embedded image (c) temporal attack added image (d) Extracted secret image.

Wavelet: The Wavelet method is utilized for watermarking using the wavelet coefficients and is utilized for decomposing the raw frame into various subbands, which include LH LL, HH, and HL. After that, the two-level decomposition is given for both the extraction and embedding processes. LSB: The LSB method is utilized for initiating the watermarking mechanisms. Here, the lesser bits of each pixel in the raw image are restored by the message bits for obtaining the embed image.

In the extraction stage, the receiver rectifies the hiding message by performing a subtraction between the pixel values of the embedded and raw image. LSB+Cost: The LSB+Cost method uses the cost values for each pixel using different criterions. The cost value is used for the extraction and the embedding process. Wavelet+Cost: The Wavelet+Cost uses wavelet method along with the cost function for video watermarking. Here, the wavelet

A.M.U. Wagdarikar and R.K. Senapati / Journal of Information Security and Applications 49 (2019) 102393

9

Fig. 5. Comparative analysis using number of frames (a) Correlation –Video 1 (b) Correlation- Video 2 (c) Correlation –Video 3.

coefficient is obtained with the use of wavelet transform and the extraction of the hidden message is done at the retrieval side. 4.3. Experimental results The experimental results of video watermarking using developed Wavelet + Chronological-MS and cost function for video 1 are depicted in Fig. 2. The original images obtained from the video frames are depicted in Fig. 2(a). In Fig. 2(b), the secret message, which is to be transmitted privacy, is depicted. The embedded image obtained after embedding the secret message with original images is depicted in Fig. 2(c). In Fig. 2(d), the resulting image of salt and pepper noise added in the embedded image for transmitting the data over a channel is depicted. Finally, Fig. 2(e), shows the extraction of the secret image from the noise added image. The experimental results of video watermarking using developed Wavelet + Chronological-MS and cost function for video 3 depicted in Fig. 3. The original image obtained from the video frames

is depicted in Fig. 3(a). The embedded image obtained after embedding the secret message with original image is depicted in Fig. 3(b). The random noise added image is depicted in Fig. 3(c). Finally, Fig. 3(d), shows the extracted secret image from the noise added image. The experimental results of video watermarking using developed Wavelet + Chronological-MS and cost function for temporal attack added video are depicted in Fig. 4. The original image obtained from the video frames is depicted in Fig. 4(a). The embedded image obtained after embedding the secret message with original image is depicted in Fig. 4(b). The noise added image is depicted in Fig. 4(c). Finally, Fig. 4(d), shows the extracted secret image from the noise added image. 4.4. Comparative analysis The comparative analysis performed by analyzing the performance of the existing techniques, like Wavelet method [29],

10

A.M.U. Wagdarikar and R.K. Senapati / Journal of Information Security and Applications 49 (2019) 102393

Fig. 6. Comparative analysis using number of frames (a) PSNR – Video 1 (b) PSNR – Video 2 (c) PSNR – Video 3.

LSB method [30], LSB+Cost (Applied LSB [30] using the cost function devised in [31]), and Wavelet+Cost (Applied Wavelet [29] using the cost function devised in [31]), with the proposed Wavelet + Chronological MS method is demonstrated in this section. 4.4.1. Comparative analysis using the number of frames The comparative analysis of proposed Wavelet + ChronologicalMS with the existing Wavelet method, LSB method, LSB+Cost method, Wavelet+ cost method using a number of frames is presented in this section. The analysis in terms of correlation coefficient taking three videos is depicted in Fig. 5. The analysis of proposed Wavelet + Chronological-MS based on correlation coefficient by varying the number of frames is depicted in Fig. 5(a). The correlation coefficient values measured by the existing methods, such as Wavelet, LSB, LSB+Cost, Wavelet+ cost, and the proposed Wavelet + Chronological MS are 0.811, 0.5, 0.477, 1, and 1

when 8 video frames are used. Likewise, when using 7 video frames, the correlation coefficient values measured by the existing methods, such as Wavelet, LSB, LSB+Cost, Wavelet+ cost, and the proposed Wavelet + Chronological MS are 0.811, 0.505, 0.465, 1, and 1, respectively. The correlation coefficient of the proposed Wavelet + Chronological-MS is 1, which improves the performance of watermarking. Fig. 5(b) depicts the performance analysis for video 2 in terms of the correlation coefficient. The correlation coefficient values measured by the existing methods, such as Wavelet, LSB, LSB+Cost, Wavelet+ cost, and the proposed Wavelet + Chronological MS are 0.811, 0.587, 0.588, 1, and 1, when utilizing one frame. Similarly, the correlation coefficient values measured by the existing methods, such as Wavelet, LSB, LSB+Cost, Wavelet+ cost method, and the proposed Wavelet + Chronological MS are 0.811, 0.587, 0.588, 1, and 1, for four frames. From the analysis, it is evaluated that the proposed Wavelet + ChronologicalMS acquires high correlation with value 1 when the number

A.M.U. Wagdarikar and R.K. Senapati / Journal of Information Security and Applications 49 (2019) 102393

11

Fig. 7. Analysis based on Salt and Pepper Noise (a) Correlation – Video 1 (b) Correlation – Video 2 (c) Correlation – Video 3.

of video frames is increased. Fig. 5(c) shows the performance analysis for video 3 in terms of the correlation coefficient. The correlation coefficient values measured by the existing methods, such as Wavelet, LSB, LSB+Cost, Wavelet+ cost, and the proposed Wavelet + Chronological MS are 0.9231, 0.3945, 0.0769, 0.9402, and 0.9594 when utilizing eight frames. From the analysis, it is evaluated that the proposed Wavelet + Chronological-MS acquires high correlation values. The analysis in terms of PSNR taking three videos is depicted in Fig. 6. Fig. 6(a) illustrates the analysis based on PSNR values for video 1. When using five frames, the PSNR values measured by Wavelet, LSB, LSB+Cost, Wavelet+ cost, and the proposed Wavelet + Chronological MS methods are 63.471 dB, 60.208 dB, 50.656 dB, 72.415 dB, and 73.126 dB respectively. Likewise, when using six frames, the PSNR values measured by Wavelet, LSB, LSB+Cost, Wavelet+ cost, and the proposed Wavelet + Chronological MS methods are 63.587 dB, 60.237 dB, 50.564 dB, 72.126 dB, and 73.018 dB respectively. Fig. 6(b) presents the analysis based on PSNR values for video 2. When using eight

frames, the PSNR values measured by Wavelet, LSB, LSB+Cost, Wavelet+ cost method, and the proposed Wavelet + Chronological MS methods are 67.998 dB, 61.019 dB, 55.106 dB, 98.190 dB, and 98.741 dB respectively. Likewise, when using seven frames, the PSNR values measured by Wavelet, LSB, LSB+Cost, Wavelet+ cost, and proposed Wavelet + Chronological MS methods are 67.647 dB, 61.019 dB, 55.135 dB, 95.179 dB, and 95.731 dB respectively. Fig. 6(c) presents the analysis based on PSNR values for video 3. When using eight frames, the PSNR values measured by the existing methods, such as Wavelet, LSB, LSB+Cost, Wavelet+ cost method, and the proposed Wavelet + Chronological MS are 65.3594 dB, 59.3405 dB, 50.3236 dB, 70.8262 dB, and 70.9073 dB, respectively. From the analysis, it can be shown that the proposed Wavelet + Chronological MS has a high PSNR value than the existing methods. 4.4.2. Comparative analysis using salt and pepper noise The comparative analysis using three input videos is depicted in Fig. 7. The varying noise levels are taken into consideration for

12

A.M.U. Wagdarikar and R.K. Senapati / Journal of Information Security and Applications 49 (2019) 102393

Fig. 8. Robustness analysis based on (a) Correlation – Video 1 using random noise (b) Correlation – Video 2 using random noise (c) Correlation – Video 3 using random noise.

analyzing the correlation coefficient and PSNR. The comparative analysis with correlation using noise for video 1 is demonstrated in Fig. 7(a). When, the noise level is 0.009, the correlation values measured by the existing methods, such as Wavelet, LSB, LSB+Cost, Wavelet+ cost, and the proposed Wavelet + Chronological MS are 0.411, 0.502, 0.465, 0.784, and 0.949, respectively. Similarly, when the noise level is 0.007, the correlation values measured by the existing methods, such as Wavelet, LSB, LSB+Cost, Wavelet+ cost, and the proposed Wavelet + Chronological MS are 0.411, 0.502, 0.465, 0.795, and 0.952, respectively. Fig. 7(b) illustrates the robustness analysis using video 2. When the noise level is 0.006, the correlation values measured by the existing Wavelet method is 0.416, LSB method is 0.587, LSB+Cost method is 0.588, Wavelet+ cost method is 0.972, and the proposed Wavelet + Chronological MS method is 0.976. Similarly, when the noise level is 0.007, the correlation values measured by the existing methods, such as Wavelet, LSB, LSB+Cost, Wavelet+ cost, and the proposed

Wavelet + Chronological MS are 0.414, 0.587, 0.588, 0.968, and 0.973 respectively. Fig. 7(c) illustrates the robustness analysis using video 3. When the noise level is 0.009, the correlation values measured by the existing methods, such as Wavelet, LSB, LSB+Cost, Wavelet+ cost, and the proposed Wavelet + Chronological MS are 0.9231, 0.3950, 0.0769, 0.9342, and 0.9539, respectively. From Fig. 7, the proposed Wavelet + Chronological-MS attain the highest values when compared to the existing methods. 4.4.3. Comparative analysis using histogram equalization and random noise Fig. 8 illustrates the robustness analysis using random noise. Here, the correlation is computed from the embedded frame and the original frame image. Fig. 8(a) depicts the robustness based analysis in terms of correlation using input video 1. When the value of random noise level is six, the corresponding correlation values measured by the existing methods,

A.M.U. Wagdarikar and R.K. Senapati / Journal of Information Security and Applications 49 (2019) 102393

13

Fig. 9. Robustness analysis based on (a) Correlation – Video 1 using histogram equalization (b) Correlation – Video 2 using histogram equalization (c) Correlation – Video 3 using histogram equalization.

such as Wavelet, LSB, LSB+Cost, Wavelet+ cost, and the proposed Wavelet + Chronological MS are 0.413, 0.502, 0.465, 0.948, and 0.996 respectively. Moreover, the analysis based on random noise based on correlation using video 2 is depicted in Fig. 8(b). When the value of random noise level is eight, the corresponding correlation values measured by the existing methods, such as Wavelet, LSB, LSB+Cost, Wavelet+ cost, and proposed Wavelet + Chronological MS are 0.411, 0.502, 0.465, 0.930, and 0.992 respectively. The analysis based on random noise based on correlation using video 3 is depicted in Fig. 8(c). When the value of random noise level is nine, the corresponding correlation values measured by the existing methods, such as Wavelet, LSB, LSB+Cost, Wavelet+ cost, and proposed Wavelet + Chronological MS are 0.9231, 0.3950, 0.0769, 0.9904, and 0.9894, respectively. From Fig. 8, it can be shown that the proposed Wavelet + ChronologicalMS obtains higher correlation coefficient value when compared to the existing methods.

The analysis in terms of histogram equalization using video 1, video 2, and video 3 is depicted in Fig. 9(a)–(c), respectively. Here, the histogram levels are varied for estimating the values of the correlation coefficient for providing improved robustness. The histogram equalization is employed for contrast enhancement by varying the intensities of the pixel. In Fig. 9(a), the correlation coefficient based on histogram equalization for video 1 is depicted. When the histogram equalization is 254, the values of correlation measured by the existing methods, such as Wavelet, LSB, LSB+Cost, Wavelet+ cost, and the proposed Wavelet + Chronological MS are 0.411, 0.502, 0.465, 0.695, and 0.962, respectively. In Fig. 9(b), the correlation coefficient based on the histogram equalization for video 2 is depicted. When, the histogram equalization is 253, the correlation values measured by the existing methods, such as Wavelet, LSB, LSB+Cost, Wavelet+ cost, and the proposed Wavelet + Chronological MS are 0.411, 0.502, 0.465, 0.696, and 0.962, respectively. The correlation coefficient based on the

14

A.M.U. Wagdarikar and R.K. Senapati / Journal of Information Security and Applications 49 (2019) 102393

Fig. 10. Robustness analysis using rotational attack (a) Correlation – Video 1 (b) Correlation – Video 2 (c) Correlation – Video 3.

histogram equalization for video 3 is depicted in Fig. 9(c). When, the histogram equalization is 254, the correlation values measured by the existing methods, such as Wavelet, LSB, LSB+Cost, Wavelet+ cost, and the proposed Wavelet + Chronological MS are 0.9231, 0.3950, 0.0769, 0.9262, and 0.9596, respectively. Finally, the proposed Wavelet + Chronological MS achieve higher values for the correlation coefficient that enhances the robustness of the watermarking technique. 4.4.4. Comparative analysis using rotational attack The comparative analysis using rotational attack by varying the rotation degrees is depicted in Fig. 10. Here, the analysis based on correlation for video 1 is shown in Fig. 10(a), the analysis based on correlation for video 2 is demonstrated in Fig. 10(b), and the analysis based on correlation for video 3 is demonstrated in Fig. 10(c). In Fig. 10(a), when the rotation degree is 5, the corresponding values measured by the existing Wavelet, LSB, LSB+Cost, Wavelet+ cost, and the proposed Wavelet + Chronological MS are 0.878, 0.500, 0.723, 0.940, and 0.947 respectively. The robust-

ness analysis based on correlation using video 2 is depicted in Fig. 10(b). When the rotation degree is 5, the corresponding values measured by the existing Wavelet, LSB, LSB+Cost, Wavelet+ cost, and the proposed Wavelet + Chronological MS methods are 0.464, 0.587, 0.588, 1, and 1. The robustness analysis based on correlation using video 3 is depicted in Fig. 10(c). When the rotation degree is 5, the corresponding values measured by the existing Wavelet, LSB, LSB+Cost, Wavelet+ cost, and the proposed Wavelet + Chronological MS methods are 0.73, 0.2358, 0.0979, 0.6797, and 0.8432, respectively. The performance of the existing and developed method is decreased when the rotation degree is improved. However, the proposed method shows the highest performance than the existing methods. 4.4.5. Comparative analysis using temporal synchronization attack The comparative analysis using temporal synchronization attack by varying the temporal changes degrees is depicted in Fig. 11. The analysis based on correlation for video 1 is shown in Fig. 11(a). In Fig. 11(a), when the temporal change is 20°,

A.M.U. Wagdarikar and R.K. Senapati / Journal of Information Security and Applications 49 (2019) 102393

15

Fig. 11. Robustness analysis using temporal synchronization attack (a) Correlation – Video 1 (b) Correlation – Video 2 (c) Correlation – Video 3.

the corresponding values measured by the existing methods, such as Wavelet, LSB, LSB+Cost, Wavelet+ cost, and the proposed Wavelet + Chronological MS are 0.8851, 0.5113, 0.7273, 0.8584, and 0.9053, respectively. The robustness analysis based on correlation using video 2 is depicted in Fig. 11(b). When the temporal change is 20°, the corresponding values measured by the existing methods, such as Wavelet, LSB, LSB+Cost, Wavelet+ cost, and the proposed Wavelet + Chronological MS methods are 0.4737, 0.5892, 0.5952, 0.8775, and 0.9522, respectively. The robustness analysis based on correlation using video 3 is depicted in Fig. 11(c). When the temporal change is 20°, the corresponding values measured by the existing methods, such as Wavelet, LSB, LSB+Cost, Wavelet+ cost, and the proposed Wavelet + Chronological MS methods are 0.8797, 0.395, 0.2769, 0.9097, and 0.9121, respectively. 4.5. Comparative discussion Here, the analysis has been provided based on the best performance obtained by the comparative methods. Table 1 shows the

comparative discussion of various techniques in terms of PSNR and correlation coefficient for video 1 and video 2. The table illustrates that the proposed Wavelet + Chronological MS method shows superior results than the existing techniques. The comparative discussion using video 3 is depicted in Table 2. Here also, the proposed method has the maximum correlation and PSNR values than the existing methods, which shows the effectiveness of the proposed method. While analyzing the performance of the comparative methods, the proposed method has the best performance than the existing methods. The reasons for the high performance of the proposed method are described below: The proposed watermarking is the transform domain based approach, which is stable, robust, and provides more imperceptibility than spatial domain-based approaches. Also, in the proposed method, the optimal regions are selected by the ChronologicalMS algorithm based on several parameters, in which the secret data has been embedded. Hence, the proposed method is more restricted to attacks. In this work, the data has been embedded by

16

A.M.U. Wagdarikar and R.K. Senapati / Journal of Information Security and Applications 49 (2019) 102393 Table 1 Comparative Discussion. Methods

Correlation coefficient Frames

Wavelet LSB LSB+cost Wavelet +Cost Proposed Wavelet + Chronological MS

PSNR (dB) Noise

Random Noise

Histogram

Frames

Video 1

Video 2

Video 1

Video 2

Video 1

Video 2

Video 1

Video 2

Video 1

Video 2

0.811 0.5 0.477 1 1

0.811 0.587 0.588 1 1

0.411 0.502 0.465 0.784 0.949

0.413 0.587 0.588 0.960 0.964

0.411 0.502 0.465 0.695 0.990

0.589 0.587 0.588 1 1

0.411 0.502 0.465 0.695 0.962

0.446 0.587 0.588 0.896 0.902

63.878 60.172 50.693 72.700 73.008

67.998 61.019 55.106 98.190 98.741

Table 2 Comparative Discussion using video 3. Methods

Wavelet LSB LSB+cost Wavelet +Cost Proposed Wavelet + Chronological MS

Correlation coefficient

PSNR (dB)

Frames

Noise

Random Noise

Histogram

Frames

0.9231 0.3959 0.0769 0.9486 0.9594

0.9231 0.395 0.0769 0.9475 0.9669

0.9231 0.395 0.0769 0.9941 0.9941

0.9231 0.395 0.0769 0.9287 0.9596

65.4717 59.3547 50.3285 71.1315 71.8421

the wavelet transform, which offers excellent spatial localization and multi-resolution characteristics, which are similar to the theoretical models of the human visual system. 5. Conclusion In this research work, a technique for video watermarking using interesting regions is designed by an optimization technique, named Wavelet + Chronological-MS. The proposed ChronologicalMS, which is developed by incorporating a chronological concept in MS algorithm, is adapted to generate an optimal region for the embedding. The proposed algorithm uses a cost function that uses several parameters, which include edge, energy, brightness, coverage, pixel intensity. After that, the two-level decomposition of the wavelet transform is given to the selected region for obtaining the wavelet coefficients for embedding. Here, the bit plane approach is employed for obtaining the binary image of the private message. Similarly, the binary hide message is embedded into the wavelet coefficient of the input frame to create the watermarked image. Then, the secret message is recovered using the cost values. The performance of the proposed Wavelet + Chronological-MS is analyzed based on evaluation metrics, correlation coefficient, and PSNR, showing superior performance with value 1.00 for correlation coefficient and 98.741 dB for PSNR. Declaration of Competing Interest None. References [1] Chen W, Li X, Zhan S, Niu D. Multimedia video watermarking algorithm using SVD and secret sharing. In: proceedings of Electronic and Automation Control Conference on Advanced Information Management, Communicates; May 2018. p. 1682–6. [2] Zhao J. Applying digital watermarking techniques to online multimedia commerce. In: Proceedings of International Conference on Imaging Science, Systems and Applications, 7; June 1997. [3] Sake A, Tirumala R. Bi-orthogonal wavelet transform based video watermarking using optimization techniques. In: Materials Today, 5; 2018. p. 1470–7. [4] Upadhyay J, Mishra B, Patel P. A modified approach of video watermarking using DWT-BP based LSB algorithm. In: proceedings of International Conference on Information Communication Instrumentation and Control; August 2017. p. 1–6. [5] Bhattacharya S, Chattopadhyay T, Pal A. A survey on different video watermarking techniques and comparative analysis with reference to H. 264/AVC. In: Consumer Electronics; June 2006. p. 1–6.

[6] Lu CS, Liao HYM. “Multipurpose watermarking for image authentication and protection. IEEE Trans Image Process 2001;10(10):1579–92. [7] Lu CS, Liao HYM, Chen LH. Multipurpose audio watermarking. In: Proceedings of 15th International Conference Pattern Recognition, 3; 20 0 0. p. 282–5. [8] Lu CS, Huang SK, Sze CJ, Liao HYM. Cocktail watermarking for digital image protection. IEEE Trans Multimed 20 0 0;2(4):209–24. [9] Lu ZM, Sun SH. Digital image watermarking technique based on vector quantisation. Electron Lett 20 0 0;36(4):303–5. [10] Chang CC, Tsai P, Lin CC. SVD-based digital image watermarking scheme. Pattern Recogn Lett 2005;26(10):1577–86. [11] Lu ZM, Zheng WM, Pan JS, Sun Z. Multipurpose image watermarking method based on mean-removed vector quantization. J Inform Assur Secur 2006;1(1):33–42. [12] Hartung F, Kutter M. Multimedia watermarking techniques. In: proceedings of the IEEE, 87; 1999. p. 1079–107. [13] Chan PW, Lyu MR, Chin RT. A novel scheme for hybrid digital video watermarking: approach, evaluation and experimentation. IEEE Trans Circu 2005;15(12):1638–49. [14] Hartung F, Girod B. Watermarking of uncompressed and compressed video. Signal Process May 1998;66(3):283–301. [15] Loan NA, Hurrah NN, Parah SA, Lee JW, Sheikh JA, Bhat GM. Secure and robust digital image watermarking using coefficient differencing and chaotic encryption, 6. IEEE; 2018. p. 19876–97. Access. [16] Shukla D, Sharma M. A new approach for scene-based digital video watermarking using discrete wavelet transforms. Int J Adv Appl Sci 2018;5(2):148–60. [17] Naseem MT, Nadeem M, Qureshi IM, Hussain A. Optimal secure information using digital watermarking and fuzzy rule base. Multimed Tools Appl 2018;, 78 , 6 :1–22. [18] Thongkor K, Amornraksa T, Delp EJ. Digital watermarking for camera-captured images based on just noticeable distortion and Wiener filtering. J Vis Commun Image Represent 2018;53:146–60. [19] Koz A, Alatan AA. Oblivious spatio-temporal watermarking of digital video by exploiting the human visual system. Circu Syst Video Technol 2008;18(3):326–37. [20] Surachat K, Amornraksa T. Pixel-wise based digital watermarking using Weiner filter in chrominance channel. In: Proceedings of International Symposium on Communications and Information Technology; 2009. p. 887–92. [21] Mabtoul S, IbnElhaj Ehassan, Aboutajdine D. Robust colour image watermarking based on singular value decomposition and dual tree complex wavelet transform. In: Proceedings of International Conference on Electronics, Circuits and Systems; 2007. p. 534–7. [22] Huang H-Y, Yang C-H, Hsu W-H. A video watermarking technique based on pseudo-3-d dct and quantization index modulation. Inform Foren Secu 2010;5(4):625–37. [23] Bassil Y. Image steganography based on a parameterized Canny edge detection algorithm. Int J Comput Appl 2012;60(4):0975–8887. [24] Ritika R, Kaur S. Contrast enhancement techniques for images–a visual analysis. Int J Comput Appl 2013;64(17):20–5. [25] Wang GG. Moth search algorithm: a bio-inspired metaheuristic algorithm for global optimization problems. Memetic Comput 2016;10(2):1–14. [26] Wagdarikar AMU, Senapati RK. Design and development of a multiobjective cost function for robust video watermarking using wavelet transform. J. Intell. Syst. 2017;28(5). [27] Porwal P, Chaturvedi RN, Ghag T, Poddar N, Tawde A. Digital video watermarking using least significant bit (LSB) technique. Int J Eng Res Technol 2014;3(1):2491–6.

A.M.U. Wagdarikar and R.K. Senapati / Journal of Information Security and Applications 49 (2019) 102393 [28] Barni M, Bartolini F, Piva A. Improved wavelet based watermarking through pixel-wise masking. IEEE Trans Image Process 2001;10(, 5 ):783–91. [29] Kumar R, Chaudhary S. Video watermarking using wavelet transformation. Int J Comput Trend Technol (IJCTT) 2013;4(5):1154–7. [30] Porwal P, Chaturvedi RN, Ghag T, Poddar N, Tawde A. Digital video watermarking using least significant bit (LSB) technique. Int J Eng Res Technol 2014;3(1):2491–6. [31] Wagdarikar AMU, K R. Design and development of a multiobjective cost function for robust video watermarking using wavelet transform. J Intell Syst 2017;28(5):873–91. [32] Asikuzzaman M, Pickering MR. An overview of digital video watermarking. In: IEEE Transactions on Circuits and Systems for Video Technology, 28; 2018. p. 2131–53. [33] UCF-Crime dataset, available at, https://www.kaggle.com/mission-ai/ crimeucfdataset. Accessed on 20-07-2018 [34] Bhopale SD, Kumar S, Jayadevappa D. Implementation of image segmentation using FPGA. Int J Advanc Res Technol 2014;3(4):2700–2. [35] Sudhkar R, Letitia S. Motion estimation scheme for video coding using hybrid discrete cosine transform and modified unsymmetrical-cross multi hexagon grid search algorithm. Middle-East J Sci Res 2015;23:848–55. [36] Ingle RB, More NS. Energy-aware vm migration using Dragonfly-Crow optimization and support vector regression model in cloud data. Int J Model Simul Sci Comput 2018. [37] Ruikar S, Chandra Prakash V, Ghuge CA. Query-Specific distance and hybrid tracking model for video object retrieval. J Intell Syst 2016;27:195–212.

17

[38] Kingsbury N. Image processing with complex wavelets. Phil. Trans. Royal Society London A 1999;357:2543–60. [39] Coria LE, Pickering MR, Nasiopoulos P, Ward RK. A video watermarking scheme based on the dual-tree complex wavelet transform. IEEE Trans Inform Foren Secur Sep. 2008;3(3):466–74. [40] Dawei Z, Guanrong C, Wenbo L. A chaos-based robust waveletdomain watermarking algorithm. Chaos, Solitons & Fractals Oct. 2004;22(1):47–54. [41] Lu CS, Chen JR, Fan KC. Real-time frame-dependent video watermarking in vlc domain. Signal Process Image Commun Aug. 2005;20(7):624–42. [42] Lee MJ, Im DH, Lee HY, Kim KS, Lee HK. Realtime video watermarking system on the compressed domain for highdefinition video contents: practical issues. Digit Signal Process Jan. 2012;22(1):190–8. [43] Huang HY, Yang CH, Hsu WH. A video watermarking technique based on pseudo-3-D dct and quantization index modulation. IEEE Trans Inform Forens Secur Dec. 2010;5(4):625–37. [44] Kong W, Yang B, Wu D, Niu X. SVD based blind video watermarking algorithm. Int Conf Innov Comput Inform Control Aug. 2006;1:265–8. [45] Liu R, Tan T. An SVD-based watermarking scheme for protecting rightful ownership. IEEE Trans Multimed Mar. 2002;4(1):121–8. [46] Shetty MV, Kumar S, Jayadevappa D. Development of brain mri image segmentation algorithms using deformable models. In: Proceedings of IEEE 3rd International Conference on Communication and Electronics Systems; 2019. [47] Kumar MVSS, Jayadevappa D. "A novel approach for segmentation and classification of brain mr images using cluster deformable based fusion approach. Period Eng Nat Sci 2018;6(2):237–42.