Mobile and Intelligent Sensing for Video Watermarking Using Spectral Centroid and Haar Wavelet Transformation on High Performace Computing

Some technologies are technologically advanced to provide security from illegal copying. Two complementary methods are encryption and watermarking. Encryption safeguards the information throughout the communication from the sender to the receiver. The data might present a distorted image after receipt and subsequent decryption. Watermarking complements encryption through embedding data openly into the image. Therefore, the watermark continuously remains existing in the data. A digital watermark is a category of indication secretly entrenched in a noise-tolerant signal similar to audio or else image information. It is indeed applied to distinguish copyright possession of such signal. Computer-aided hiding of the given digitized information in a carrier is known as watermarking. Digital watermarks possibly will be employed to validate the authenticity or integrity of a carrier signal or to determine source uniqueness. It is evidently applied for determining copyright contraventions and aimed at banknote verification. Analogous to traditional watermarks, digital watermarks are unique only beneath certain conditions. Once a digital watermark varies a carrier in a manner that it turns out to be noticeable, formerly it is of no use. The media will be visible by traditional watermarks (similar to images or else video) but the signal might be pictures, video, audio, texts or 3D models in digital watermarking. A signal can transmit some different watermarks at the equivalent time. Image watermarking is achieved in this study using two methods known as Hidden Markov Tree– Contourlet Wavelet Transform (HMT-CWT) and Haar wavelet transform – Discrete Fourier transform (HWT-DFT). In the next HWT-DFT method, a video is given as an input and it is split into two halves (audio and image). The audio is de-watermarked through Spectral Centroid Wavelet Transform and enhanced by utilizing Firefly procedure. The images is handled through HWT in addition to DFT. Then the output watermarked images and audio combined together to form a watermarked video. The obtained video is de-watermarked to produce the original copy of the video. The process of getting back the original copy by removing the watermark from the video is called as de-watermarking. two levels: applicative integration and ground-truth confrontation. The validation of ground truth was built on two criteria, Precision and Discriminance.


Introduction
The previous a long time have seen an outstanding progression in virtual communication and computer era. The blessings they provide are an green transmission of facts, high computing power, availability of excessive bandwidth, clean enhancing of virtual contents, etc.
It is now clean to copy virtual content with none loss inside the great of the virtual media. The boom in this era has additionally conveyed unique issues besides its benefits. The hassle of ease to copy virtual content material, edit, and transfer, it is able to violate copyright protection. Digital watermarking has been advised as an advanced approach to affirm possession of virtual data (Katariya 2012).
Watermark evidence is entrenched into the original information (crowd indication) in such a way that it stays existing on circumstance that the clear exceptional of the content of host sign is at an appropriate level. The original records owner demonstrates his/her possession by using extracting the watermark statistics from the watermarked content (watermarked signal) in case of multiple ownership claims.
The strategies used for information protection are cryptography, steganography and watermarking.
Cryptography offers with securing contents of a message (host signal), after encryption, the message seems like noise and is useless. However, encryption structures do not completely solve the security hassle; due to the fact once decryption is performed.
There is not any control over the dissemination of the facts ( Barni andBartolini 2004 andLu 2004).
Hence, a method is required which continually offer safety to host signal. Steganography and watermarking are seemed as the techniques which give protection to host sign all the time. The basic idea of steganography is to embed a secret message in a media that is interpreted through the supposed user only.
Steganography pacts with hiding a message, however watermarking have a further obligation of securely hiding the message, in such a way that the message cannot interfere. A right Steganographic scheme can entrench a massive amount of data without a perceptual degradation to multimedia statistics.
On the alternative hand, a great watermarking scheme would entrench records that couldn't be changed or eliminated without growing the multimedia signal unusable. A watermarking gadget involves a tradeoff among imperceptibility, robustness and embedding capacity (Barni and Bartolini 2004). It includes steps, viz. Watermark embedding and watermark detection/extraction. An encryption key is a mystery that is used in cryptography to make hidden information more secure.
The watermark records may be a random number that is generated the use of a mystery key which is known as the seed of watermark. Embedding is a process with the aid of which watermark records is fused with the original signal (additionally referred to as host signal) to offer a watermarked signal.
Detection manner tests the presence of watermark data within the watermarked sign, however the extraction system unearths the watermark records from the watermarked signal. User keys are mystery which is used for watermark embedding. These keys can be watermark length, watermark location, watermark scaling factor, etc.
Digital Watermarking is the phenomena of hiding/embedding digital statistics right into a host sign in such a way that sign keeps its perceptual transparency (Bender et al 1996 Abraham et al (2016) proposed a paper on set of rules for embedding a photograph or logo in a colored photo. They endorsed two mask to provide less distracting to eyes of the human as embedding and repayment mask. In this approach, the information of the watermark was widely spread to enhance its stability against the attacks. They highlighted that the occurrence of S and P noise attack when the pixel was enhanced due to its nearby pixel which would be resolved by their methodology.
The implementation of mask aided them to have an even distribution of the modification that was carried out. The suggested algorithm was verified on many images. The imperceptibility and robustness of the The optimum sub-watermarks are selected from the inserted watermark and then it combined with the final watermark to conglomerate the sub-watermarks. The suggested algorithm has robustness beside geometric attacks and standard image processing was investigated from the outcomes.
Pardhuet al (2016) concentrated on DWT and DCT frequency domain techniques and they successfully embedded and extracted the watermark into the image of the cover and showed no perceptual variation among the watermarked image and cover image, which fulfilled one of the chief necessities of the digital watermarking, i.e., imperceptibility.
In the application, diverse watermarked and cover images were considered. They calculated the SNR values for each image and stated that when the SNR increases the DATA for the images also increased. In their research, watermark and cover images were of the same size and if the watermark size was fewer than the image of the cover, then improved outcomes could be done to endure the watermarked image during image processing. They demonstrated that their watermarking plan was strong to normal flag preparing attacks and suggested strategy accomplished esteems of PSNR from 13 dB to 24 dB for several watermarked sounds.
These results showed that their suggested strategy of watermarking could be a sensible competitor for copyright security of sound.
Sk et al (2018) developed the scheme for the protection of data in IoT applications. They stated that existing wireless technologies has no authentication of information and proprietorship despite providing data secrecy through encryption mechanisms. Thus the privacy and security of the data on the IoT applications were of core concern. This scheme was intended at providing ownership and data authentication by Play fair cipher variability.
The false positive singular value problem of breakdown based watermarking has received the resolution from the scheme. The suggested scheme was outclassing the contending methods concerning top signalto-noise proportion, minute fault rate correlation, and also evading false positives which shown from the experimental results.

Transform (HWT-DFT)
In the proposed method, a video is given as an input and it is split into two halves, one is audio and the other is image. The audio and image watermarking process is achieved by means of utilizing two techniques. The process of audio watermarking can be done by using spectral centroid wavelet and firefly algorithm. Whereas the process of image watermarking can be achieved through discrete Fourier transform and Haar wavelet transforms.

Figure 1 Watermarking Process
Firstly, the incoming video is separated into two halves. The first half is watermarked as an audio signal whereas the second half is watermarked as an image. The above-mentioned figure 5.1 represents the chunk drawing of the future architecture. Here, the secret information is hidden as audio and an image.
The security is achieved through this process.

Audio and Image Watermarking Using HWT-DFT
In the proposed research work, audio and image watermarking using spectral centroid wavelet and firefly algorithm. The incoming video is fragmented into two halves. The primary one is watermarked as an audio signal then the next one is watermarked as an image.
 Audio signal-Processed using Spectral Centroid Wavelet transform.
 Another audio Signal-Taken and watermarked with the input audio signal and optimized with the firefly algorithm.
 The image is processed through the Discrete Fourier Transform and compressed further by considering the high coefficient using the Haar wavelet transforms.
 Another input for image watermark is taken and combined with the previous output of image pre-processed using the embedded OOPM algorithm.
 The watermarked audio and image are combined to form watermarked video The flow diagram of the proposed methodology is described below-given Figure 5.2.

Audio Watermarking
The audio watermarking process mainly consists of four stages namely, The received audio signal is decomposed into several bands. The extraction of the feature of the audio signal is achieved through Spectral Centroid Wavelet (SCT) transform. It offers the center of gravity of the magnitude spectrum. It differs the sinusoidal wave of a signal and further defines the specific location on the sub band. Spectral Centroid Frequency (SCF) assesses the center of gravity of a spectrum in every single sub-band. SCF is computed as the average weighted frequency of the sub band as given in below Where Xk denotes the magnitude of the component in frequency band fk.

Spectral Centroid Wavelet Transform
The spectral centroid is the median of the spectrum. Since spectrum gives the sign of how the signal's amplitude is dispersed among the frequencies, its center of mass shows the average amount of amplitude.
From the audio view, it is the average loudness, and when compared to image view, it is taken as average brightness. The weighted average is taken for calculation from the signal observed. The watermarked audio is further process through the firefly algorithm

Firefly Algorithm (FA)
In a FA calculation, the target work streamlining is unexpected at the brilliance notwithstanding a development of the firefly. The more brilliant firefly will bait the neighboring fireflies with low splendor.
The firefly procedure starts with the firefly's populace instatement. The splendor of the firefly decides the development of the fireflies. In the iterative method, the profundity of the ithfirefly is in correlation with the force of jth firefly. In view of the differentiation in force, both ith firefly draw nearer to jth firefly or jth firefly will move towards ith firefly. The most appropriate arrangement got is persistently refreshed until the specific halting condition is satisfied. When the iterative framework includes an end, the agreeable answer is resolved. Figure 5.1 portrays the pseudocode of the FA calculation. Pseudocode: Begin; Initialize algorithm parameters: Define

Haar Wavelet Transform
Haar wavelet basis may be utilized to illustrate a picture by a wavelet rework computation. The pixel is averaged collectively pair-smart and is determined to collect the new resolution photo via pixel values.
Particular records will be lost in the averaging procedure. The Haar wavelet rework is applied to investigate snap shots successfully and efficaciously at several resolutions. It is utilized to gather the approximation coefficients similarly to element coefficients at exclusive levels. The Haar remodel works like a low-skip clear out as well as a high-bypass filter out at the equal time.Haar wavelet function, represented by (t), is specified by

Discrete Fourier transforms
Image compression techniques such as pixel coding, predictive coding, and transform coding. The Given F(u,  ), it can obtain f(x, y) back by means of the inverse DFT and given by, The values of F(u,  ) in the above equations are called the Fourier coefficients. For u=0,  =0, we have, f(x, y) is the entire leaden gage equal of the copy also called the dc constituent of the Fourier alter also represent zero frequency.
Even if f(x, y) is real, its transform is complex. If R(u, ) and I(u, ) represent the real and imaginary components of F(u, ), the DFT can be expressed in polar form, Thus, an image obtained by taking the inverse Fourier transform is also infinitely periodic. But DFT implementations compute only one period, so we work with arrays of size M x N.
The Fourier transform pair satisfies the following translation properties, And That is multiplying f(x, y) by the exponential shifts the origin of the DFT to = , u = and conversely, multiplying F(u,  ) by the negative of that exponential shifts the origin of f(x, y) to (x0 , y0 ).

Ordered Orthogonal Matching Pursuit Algorithm
The OOMP is enhanced version of OMP procedure that notices less co-efficient in the support set is removed by associating with the projection coefficients. The procedure is scaled to evade re-entry. It achieves accurate retrieval of copy with the less quantity of measurements. Furthermore, the optimum values are transformed into binary signals for less computation of the complete image. Thus produces an embedded output of a watermarked image by means of associating input images coefficient.

De-Watermarking Process
The image output is united with a watermarked audio to produce a watermarked video. At any time if the fusion process happens, it should also be competent to acquire back the original copy. The procedure of attaining the original copy back by way of eliminating the watermark from the video is known as the De watermarking process. This procedure pursues a similar process which was followed by the watermarking process, but the only difference or dissimilarity is that this process can be implied with the different algorithms. The watermarked video is separated for the second time into audio and image. Audio is dewatermarked through the spectral centroid wavelet transform, and the Discrete Wavelet Transform remarks the image. Subsequently the removal of a watermark, the audio and image are additionally merged to produce the original copy of the video.

Audio De-Watermarking
Sound sign A * which is going to be removed is separated into M sound casings, indicated as A * 1(p), p =

Image De-Watermarking
In this procedure right off the bat DWT is applied to watermarked picture and spread picture which deteriorated the picture in sub-groups. After that the watermark is recuperated from the watermarked picture by utilizing the equation of the alpha mixing.
As per the recipe of the alpha mixing the recuperated picture is given by Can forget the lack of geometric synchronization during de watermark process.5 Results of Watermarking Using HWT-DFT.
This section discusses the results obtained from image and audio watermarking processes which are described below.

Image watermarking process
The quality of the original signal is identified by this parameter. It gives information about the similarity between the original and the watermarked audio signal. PSNR is calculated by using = 20 log 10 255 √

Figure 4 PSNR of Images
The method which has the highest PSNR value gives overall similarity between the original and    In video 1, image 1 is the original image and image 2 is the secret image to be watermarked.
Watermarked image is the last in which the secret image is watermarked in image 1. Table 1 Videos showing the images, watermark and watermarked images In video 2, picture 1 is the first picture and picture 2 is the mystery picture to be watermarked. The last picture is a watermarked picture in which the mystery picture is watermarked in picture 1.
In video 3, picture 1 is the first picture and picture 2 is the mystery picture to be watermarked. The last picture is the watermarked picture in which the mystery picture is watermarked in picture 1.
In video 4, picture 1 is the first picture and picture 2 is the mystery picture to be watermarked. The last picture is the watermarked picture in which the mystery picture is watermarked in picture 1.
In video 5, picture 1 is the first picture and picture 2 is the mystery picture to be watermarked. The last picture is the watermarked picture in which the mystery picture is watermarked in picture 1.

Mean-Squared Error (MSE)
The MSE represents the cumulative squared error between the watermarked image and the original image As shown in Figure 5.4, the Mean Square Error of different images in videos is shown. The MSE is higher for video 1 when compared with video 2 and video 3. Video 4 MSE value is somewhat more as compared to video 2, and video 5 MSE value is less than video four but more than that of video 3.

Peak Signal-To-Noise Ratio (PSNR)
The pinnacle signal-to-clamor proportion, in decibels, between two pictures. This proportion is frequently utilized as a quality estimation between the first and a watermarked picture. The higher the PSNR, the better the quality picture. To register the PSNR, the square initially figures the mean-squared blunder utilizing the previously mentioned Equation (5.12): In the past condition, M and N are the quantity of lines and sections in the info pictures, individually. At that point the square figures the PSNR utilizing the accompanying condition: In the past condition, R is the greatest variance in the information picture information type.
For instance, on the off chance that the info picture has a twofold exactness drifting point information type, at that point R is 1. On the off chance that it has a 8-piece unsigned whole number information type, R is 255, and so forth.

Normalized Cross Correlation (NCC)
The NCC gets the values in the interval [0, 1], where 1 indicates the best match.
where μx, μy, σx,σy, and σxy are the local means, standard deviations, and cross-covariance for images x, y.

Figure 15 Attack
Shows the attack in the noise signal. Figure 15.
shows the watermarked audio signal. The audio signal given as input is watermarked.   The comparison of the computational complexity is given in the  Proposed methodology 3.13

Conclusion
In this chapter, Video will be given as a input that is divided into two parts that is audio and image. The process of audio and image watermarking using HWT-DFT was explained.

.1 Conclusion And Future Scope
HWT-DFT technique, video might be given as input and its miles divided in to components audio and photograph. The Audio is processed via Spectral Centroid Wavelet Transform and optimized by the The method which has the highest PSNR value gives overall resemblance amid the unique and waterlined

FUTURE SCOPE
The future directions for this study will be as follows:

1) Removal of Multiple types of Noise
An image may be corrupted by multiple types of noise. Image fusion algorithms can be implemented so that multiple types of noise can be removed.

2) Fusion in the presence of blur
Presence of blur is another common problem of imaging application. Presence of blur in the image makes the extraction of the critical feature, difficult. Blur may occur simultaneously with different types of noise. Hence, deblur is also an important task to increase the information content and quality of an image. Therefore, deblurring should also be incorporated with image fusion both in the presence or absence of noise to increase the adaptiveness of the fusion algorithm.

3) Image Registration
In the future, work could be done on image registration, and it should be combined with the image fusion. So that, the fusion algorithms would not be limited to the registered images and can be applied to unregistered images.

4) Hybrid Algorithm
Hybrid algorithm (FA) and differential evolution (DE) called as Hybrid Firefly Algorithm (HFA) that combines the attraction mechanism of FA with the mixing ability of DE which increases the speed of convergence and the population diversity.