Robust Watermarking via Multidomain Transform Over Wireless Channel: Design and Experimental Validation

Watermarking is a technique that provides ownership rights for shared data, and this technique can also help to transmit confidential information to realize secure communication. In this paper, we design a blind digital watermarking based on Discrete Wavelet Transform-Discrete Cosine Transform-Singular Value Decomposition (DWT-DCT-SVD) and optimize it by genetic algorithm (GA). This watermarking has both high robustness and high imperceptibility. Then we design a transmitter and receiver system based on the IEEE 802.11a and software-defined radio (SDR). We adopt several optimization methods for the system, such as ACK/HARQ mechanism, different redundant coding, frame length adaption, adaptive modulation, and so on, aiming to improve communication reliability and rate. We verify its high communication reliability and rate through actual experiments and complete the combination of digital watermarking and wireless transmission. We could build the application scenario of Image Security Wireless Communication based on Wireless Local Area Network (WLAN) protocol and SDR platform indoors or outdoors in small areas.

mation theft and copyright theft incidents have increased, 23 seriously infringing the legitimate rights and interests of 24 digital product holders and becoming a major obstacle to 25 digital technology's healthy and sustainable development. 26 Under this background, information hiding technology is 27 born [1] and thrived, and digital watermarking technology is 28 The associate editor coordinating the review of this manuscript and approving it for publication was Gulistan Raja . an essential technology in information hiding technology for 29 the evidence of rightful ownership [2]. 30 In recent years, the application of digital watermarking 31 technologies has developed rapidly, including fields of pic-32 ture, video, audio, text, software, and so on. All kinds of 33 digital watermarking technology help to ensure information 34 security and copyright [3], [4]. For example, recently emerg-35 ing central bank digital currencies also use digital watermark-36 ing for secure encryption and authentication [5], which can 37 show the importance of digital watermarking technology in 38 such an era of information and network. 39 The key to digital watermarking technology is using digital 40 processing technology to embed the authentication informa-41 tion with security and confidentiality into the carrier. The 42 performance of digital watermarking mainly depends on three 43 aspects: security, robustness, and imperceptibility [6]. 44 Image Security Wireless Communication in this paper uses 45 WLAN protocol to realize secure image transmission with 46 digital watermarking, even under varying channel conditions. 47 Since then, many digital watermarks combining SVD with 105 other frequency-domain algorithms have appeared succes-106 sively. Liu and Liu proposed a digital image watermark-107 ing algorithm based on DWT and SVD in [13], which 108 has a stronger anti-attack ability than the SVD method. 109 This algorithm is robust to JEPG compression, noise, low-110 pass filtering, median filtering, contrast enhancement, and 111 other commonly used signal processing techniques. Lai 112 and Tsai proposed a hybrid image watermarking technique 113 based on DWT and SVD in [14], embedding the water-114 marking into the singular value of the subband covering 115 the wavelet transform. This technique makes full use of 116 the respective characteristics of the two transform domain 117 methods, and the space-frequency localization of DWT and 118 SVD can effectively represent the inherent algebraic prop-119 erties of images. Experimental results show that the pro-120 posed method improves invisibility and robustness under 121 attack.

122
The watermarking algorithm based on SVD does not have 123 the disadvantage that traditional watermarking technology 124 cannot resist attack. However, DCT-based watermarking pro-125 vides compression, while DWT-based compression provides 126 extensibility. Therefore, a new watermarking technique can 127 be created by combining these three transforms. Navas et al. 128 mentioned that watermarking method could be built based 129 on DWT-DCT-SVD in [15]. The watermarking information 130 is hidden in the DCT coefficients of the DWT coefficients, 131 which brings robustness to the watermarking, and the visual 132 watermark can be extracted without a reasonable amount of 133 distortion even under various attacks. Although the presented 134 work in this paper has similarities with those work based 135 on DWT-DCT-SVD in [15], [16], and [17], the watermark 136 embedding and extraction algorithm in this paper does not 137 need prior information so that we can realize digital blind 138 watermarking, such as [18].

139
With the continuous development of machine learning, 140 in addition to the frequency domain algorithm, a digital 141 watermark embedding algorithm based on the support vector 142 machine (SVM) was proposed by Li et al. in [19]. Because of 143 the excellent learning and generalization ability in address-144 ing the problem of small sample learning, SVM is a good 145 way to memorize the relationship between randomly selected 146 image pixels and adjacent pixels. At the same time, SVM 147 can adjust or compare the relationship between the embed-148 ded pixel and the output of SVM to extract the watermark. 149 The experimental results show that SVM has good visual 150 perception ability, high security, and practicability. Tsai     In view of the above challenges, with the above digital water-217 marking technology on the basis of SDR, the contributions 218 of this paper are as follows. Figure 1 shows the main design 219 framework.

220
• We design a blind digital watermarking algorithm based 221 on DWT-DCT-SVD and optimize it with genetic algo-222 rithm. The method based on DWT-DCT-SVD can make 223 the watermark more imperceptible and robust, and the 224 genetic algorithm optimizes the digital watermark by 225 calculating the most appropriate watermark embedding 226 factor to achieve a balance between robustness and 227 imperceptibility.

228
• We combine the SDR and the image watermarking pro-229 cess to realize the Image Security Wireless Transmission 230 based on IEEE 802.11a and PLUTO-SDR. We want to 231 build the application scenario for image security com-232 munication based on WLAN protocol and SDR platform 233 indoors or in a small outdoor area.

234
• We construct a transmitting and receiving experimen-235 tal platform. We optimize and improve it through 236 ACK/HARQ mechanism, different redundant coding, 237 frame length adaption, adaptive modulation, and other 238 methods. As a result, the image transmission rate is 239 increased while the reliability of communication trans-240 mission is guaranteed.

242
In this section, we will present our blind digital watermarking 243 algorithm. Firstly, we will show the process flow of water-244 mark embedding and extraction. Then, we will explain the 245 embedding and extraction algorithm in detail. Finally, we will 246 show how to utilize GA to optimize the watermarking. The formula for PSNR of an image with a size of M × N is 250 as follows, which mainly characterizes the imperceptibility of 251 the image after the digital watermark is embedded. In general, 252 when the PSNR is greater than 35dB, it is difficult for the 253 human eyes to recognize the change in the image, The formula for NC of the digital watermark with a size of 259 P×Q is as follows, which mainly characterizes the robustness 260 of the digital watermark in the attack and extraction process,

292
where W (i, j) is the pixel value in row i and column j in the 293 scrambled binary watermark image.

294
After that, we will obtain the hidden watermark without 295 any prior information. Jagadeesh et.al. proposed a method that uses the genetic 303 algorithm to optimize digital watermarking in [28]. There-304 fore, the genetic algorithm is also used in this paper to cal-305 culate the most suitable embedding factor to achieve the best 306 balance of imperceptibility and robustness.

307
The fitness function is set as, In this section, we will present our transmitter and receiver 316 system based on IEEE 802.11a [29]. Firstly, the transmitter 317 and the receiver design are shown with process flow. Then 318 we will explain how to use several optimization methods to 319 optimize the communication system to improve communica-320 tion rate and reliability.

322
The design of the transmitter is shown in Figure 4. Depending 323 on IEEE 802.11a, the transmitter converts binary bitstreams 324 to MAC service data units (MSDUs), MAC protocol data 325 units (MPDUs), and physical service data units (PSDUs) in 326 sequence and then completes the steps of the MAC layer. 327 Then the transmitter performs physical operations on the 328 data to generate physical protocol data units (PPDUs) and 329 the transmitted signal. It's noticed that the transmitter will 330 transmit signals at different coding rates and adjust the mod-331 ulation order adaptively. Details are provided in the following 332 section.

334
The design of the receiver is shown in Figure 5 accord-335 ing to IEEE 802.11a, including basic physical and MAC 336

340
The following optimization methods are adopted at the trans- Before the transmitter transmits data for the first time, 377 it will send a Request-to-Send (RTS) signal. When the 378 receiver receives the RTS signal, it will send an ACK signal 379 for confirmation to complete the handshake operation, thus 380 ensuring the transceiver synchronization of the transmitter 381 and receiver.

383
In this section, we mainly conduct three evaluation exper-384 iments: digital watermark performance evaluation, packet 385 error rate experiment, and SDR-based image communication 386 experiment. The image chosen in this experiment is Lena. tiff, and the 391 embedding factor ranges from 2 to 12.

392
From Figure 6, we can see the relation between the embed-393 ding factor and PSNR/NC. With the increase of the embed-394 ding factor, PSNR will show a peak-like trend of rising first 395 and then declining, while NC will continue to increase until 396 1. This result visually verifies the importance of using the GA 397 algorithm to find the optimal embedding factor to achieve the 398 best balance between PSNR and NC. Four images are used as the digital watermark embedding test 402 images: Lena. tiff, Peppers. tiff, Baboon. tiff, and Airplane. 403 tiff. Because the weight γ in the fitness function in the genetic 404 algorithm is different, the optimal embedding factor is differ-405 ent. For each image, 44, 46, and 48 are used as the weight γ 406 in the genetic algorithm to obtain the best embedding factor, 407 PSNR, and NC. Besides, the algorithm proposed in this paper 408 will be compared with the methods in [16], [17], and [18] on 409 PSNR and NC.

410
It can be obtained from Table 1 that the optimal watermark 411 embedding factor calculated by GA can make the PSNR of 412 VOLUME 10, 2022   Table 2 shows that compared with other algorithms, 420 the watermarking algorithm proposed in this paper has the 421 highest PSNR based on ensuring NC.

422
Unless watermarks, confidential images or binary bit-423 streams can also be embedded into images to achieve the pur-   Figure 7 shows that as the number of iterations increases, 434 the optimal fitness of each generation will continue to con-     Table 3, it can be concluded that the digital water-448 mark designed in this paper has good robustness against 449 brightness adjustment, noise attacks, cropping attacks, fil-450 tering attacks, and JPEG compression attacks. Because 451 of its excellent robustness against JPEG compression, the 452  Table 4 shows that compared with method [18], the water-   rate is 1/2 and 64QAM while the convolutional coding rate 471 is 2/3. The simulation test condition selected the Gaussian 472 channel. The result is shown in Figure 8.

473
It can be seen from the result that although the higher 474 modulation order can increase the communication rate, the 475 system will have poorer anti-interference performance as a 476 price. The coding rate will also affect the anti-interference 477 ability, and the lower coding rate means more redundant bits. 478 Therefore, the system will have a stronger anti-interference 479 ability.

480
The methods of different redundant coding and adap-481 tive modulation designed in this paper will significantly 482 ensure the balance between the communication rate and the 483 anti-interference ability of the communication. Using dif-484 ferent coding rates means that the lower coding rate will 485 ensure transmission reliability if the channel deteriorates. 486 Using adaptive modulation implies that when the channel 487 deteriorates, a lower modulation order is used to provide 488 transmission reliability. In contrast, a higher modulation order 489 is used to ensure the transmission rate when the channel 490 returns to normal. The actual experiment in this paper is based on the MATLAB 494 software platform and ADALM PLUTO hardware platform. 495 ADALM-PLUTO is an SDR hardware module designed and 496 produced by Analog Devices Inc (ADI), which is based on 497 AD9363 and provides a receiving and transmitting channel. 498 The parameters of ADALM-PLUTO are shown in [30], the 499 setting of experimental conditions is shown in Table 5, and 500 the actual experiment environment is shown in Figure 9.   The modulation order is adaptively reduced when the channel 506 condition deteriorates to ensure communication reliability.

507
When the channel condition returns to normal, the transmitter 508 and receiver restore the modulation order to provide a high 509 communication rate.  Table 6 shows the EVM and SNR tested by the receiver 512 corresponding to the five constellation diagrams in the con-513 stellation diagram experiment. When the channel condition 514 deteriorates, the EVM is higher, and the SNR is lower. When 515 the channel condition becomes normal, the EVM decreases 516 while the SNR increases.  Through the above experiments, it can be seen that even 524 under the deterioration of channel conditions, adaptive modu-525 lation and ACK/HARQ mechanism can balance the reliability 526 and the rate of the communication system so that the commu-527 nication system can obtain a high communication rate based 528 on high reliability.

530
In this paper, we design a blind digital watermarking algo-531 rithm based on DWT-DCT-SVD and optimize it by the 532 genetic algorithm. Then we verify its high imperceptibility 533 and robustness through experiments and evaluation. Because 534 of its high PSNR and the unique embedding and extraction 535 algorithm proposed in this paper, we can fully guarantee its 536 security so that confidential information can be embedded 537 into the image as the watermark for secure communication 538 and image forensics. 539 We design a transmitter and receiver system based on the 540 IEEE 802.11a standard and combine it with digital water-541 marking to achieve Image Security Wireless Communication. 542 We adopt the methods of ACK/HARQ mechanism, different 543 redundant coding, frame length adaption, adaptive modula-544 tion, image compression coding, and RTS operation to ensure 545 both high communication rate and reliability.

546
Toward future watermarking, there are still the following 547 open problems, including machine learning and deep learning 548 to realize the extraction of the watermark and fragile water-549 mark to identify a tampered picture.