Chain-Of-Evidence in Secured Surveillance Videos Using Steganography and Hashing

Home , Steganography, Surveillance

2020 IEEE Intl Conf on Dependable, Autonomic and Secure Computing, Intl Conf on Pervasive Intelligence and Computing, Intl Conf on Cloud and Big Data Computing, Intl Conf on Cyber Science and Technology Congress

Chain-of-Evidence in Secured Surveillance Videos using Steganography and Hashing

Nadia Kanwal Mamoona Naveed Asghar Mohammad Samar Ansari Software Research Institute Software Research Institute Software Research Institute Athlone Institute of Technology Athlone Institute of Technology Athlone Institute of Technology Athlone, Ireland Athlone, Ireland Athlone, Ireland [email protected] [email protected] [email protected] Lahore College for Women University The Islamia University Alighar Muslim University Lahore, Pakistan Bahawalpur, Pakistan Aligarh, India [email protected]

Brian Lee Martin Fleury Marco Herbst Yuansong Qiao Software Research Institute School of EAST Evercam Ltd. Software Research Institute Athlone Institute of Technology University of Suffolk Dublin, Ireland Athlone Institute of Technology Athlone, Ireland Ipswich, UK [email protected] Athlone, Ireland [email protected] ﬂ[email protected] [email protected]

Abstract—Video sharing from closed-circuit television video They potentially help law enforcement agencies to utilize recording or in social media interaction requires self- the resulting video recordings as proof of a crime or illicit authentication for responsible and reliable data sharing. Sim- ilarly, surveillance video recording is a powerful method of activity. However, due to equally rapid advances in the deterring unlawful activities. A Solution-by-Design can be field of image processing, surveillance data can be easily helpful in terms of making a captured video immutable, as such tampered with. Examples include [2]: regional alteration of recordings cannot become a piece of evidence until proven to be intra-frames, through cut-and-paste; and inter-frame forgery, unaltered. This paper presents a computationally inexpensive by the insertion of video frames. Furthermore, transmission method of preserving a chain-of-evidence in surveillance videos using steganography and hashing. The method conforms to the errors also contribute to the alteration of videos, if they data protection regulations which are increasingly adopted by are not suitably protected with forward error correction governments, and is applicable to network edge storage. Secu- and the like. Consequently, such recordings are not directly rity credentials are stored in a hardware wallet independently admissible in a court of law as evidence until they are proven of the video capture device itself, while evidential information to be authentic through forensic analysis [3]. Unfortunately, is stored within video frames themselves, independently of the content. The proposed method has turned out to not applying forensic analysis is not cheap in terms of time and only preserve the integrity of the stored video data but expense. Thus, an important objective for a Closed Circuit also results in very limited degradation of the video data TeleVision (CCTV) systems is that it should provide a chain- due to steganography. Despite the presence of steganographic of-evidence, according to stated rules, so that the videos can information, video frames are still available for common image prove themselves to be an authentic piece of information, processing tasks such as tracking and classification. avoiding or reducing the need for forensic analysis. Keywords-Video Security, Video Surveillance, Steganogra- A further driver towards self-authentication is the need phy, Hashing, Information Sharing for storage and analysis of huge amounts of multimedia I. INTRODUCTION data, which has propelled companies towards cloud storage for efficient access [4]. However, following on from the It is no longer the case that only special places are European Union’s ratification of the General Data Protection kept under surveillance for security and safety purposes. Regulation (GDPR) in May 2018 (and other similar data Surveillance is rapidly becoming a requirement for almost privacy regulations in other countries), cloud storage may be every house, office, and public place, when smart tracking also be unacceptable, due to the potential for unauthorized can be applied [1]. This development is fuelled by the avail- access to video data by third parties. GDPR focuses on ability of low-cost and small-sized surveillance cameras. reversible data protection [5] via encryption [6] as a data- This paper is part of the Marie Skłodowska-Curie Career-FIT Postdoc protection safeguard, which is also employed herein. This Fellowship programme under project ID: MF-2018-0058 funded by the paper presents a multifaceted solution to the problem of European Union’s Horizon2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No. 713654 and Science self-authentication and the requirements of data protection Foundation Ireland (SFI) under Grant Number SFI 16/RC/3918 and the regulations (e.g. GDPR) in respect to storage. The proposed European Regional Development Fund. solution works as follows:

978-1-7281-6609-4/20/$31.00 ©2020 IEEE 263 DOI 10.1109/DASC-PICom-CBDCom-CyberSciTech49142.2020.00053 • Captured video data are protected by calculating their Hashing functions, also referred to as message digest hash. functions, work by extracting a fixed-length bit string from • Secondly, these hashes are stored inside video frames a given message (text, image, video). Such functions have as hidden information. found varied applications in database search and compilers • Thirdly, every frame holds the hash of its previous two as well as in cryptography. Over the recent past, there has frames and, therefore, creates a chain-of-evidence. been significant interest in employing hash functions in mul- • Lastly, all this data are encrypted and stored on the timedia applications for both indexing and security. A salient network edge. feature of most cryptographic hashing algorithms such as • To further enhance the security, encryption keys are Message Digest 5 (MD5) and Secure Hash Algorithm 1 stored in a hardware wallet, separately from the device (SHA-1) is that they are overly sensitive to the contents of holding the actual video. the message. Altering even a single bit of the input changes The power of the proposed method is that it hides the the output hash completely. However, multimedia data such evidential data inside the video frames themselves, with- as digital images go through various manipulations such as out affecting the image content. Furthermore, the proposed cropping, scaling, enhancement, and compression. method makes it easier to maintain, synchronize, and prevent An image hash function therefore takes into account the loss of, evidence. Such a record of the chain-of-evidence is changes in the visual domain and produces hashes based desirable not only in the case of surveillance videos, but the on the overall visual appearance. Such a hash function same could also be very useful in ascertaining the originality is useful in identifying images in databases, in which and authenticity of videos uploaded to social media sites, the image possibly undergoes incidental changes (such as and could also be used for the identification of fake news format changes, compression, scanning, signal processing videos. operations or watermarking). Another application of image hashing is in robust image authentication wherein the hash II. BACKGROUND remains invariant under perceptually minimal alterations to As is well known, steganography refers to the hiding of the image, but detects malicious tampering of image data. information (text, audio, image, video) in another carrier The SHA256 hashing algorithm has been used in this work media (usually referred to as the ’cover’). There are four [20]. essential properties of a good steganographic system, viz. imperceptibility, security, information hiding capacity, and III. METHODOLOGY robustness. There have been several different approaches to achieve the goal of information hiding in the cover In order to create a chain-of-evidence using hashes, a media. The simplest approach works by manipulating the salted hash is first created by concatenating half of the bytes Least Significant Bits (LSB) of the three color channels of the hashes of the previous two video frames. Therefore, (RGB), in light of the fact that the LSBs carry very minimal each frame will store: a salted hash of data from the previous information. This spatial approach ensures that the overall two frames (256 bits); and other optional data such as the visual aesthetic of the image is not altered. Initial works in camera identity; the date and time of video capture; and LSB steganography concentrate on designing the system to the Global Positioning System (GPS) location. We call this increase the payload capacity by utilizing most of the cover information the evidential data. Subsequently, the frame’s image pixels [7]–[15]. However, steganalysis techniques own hash (256 bits), after the insertion of the evidential soon became strong enough to break such systems using data, is calculated and inserted. The size of the data to be statistical analysis. Therefore, sophisticated robust LSB tech- stored becomes 512 bits per frame considering salted hash niques based cryptography-steganography which can evade and current hash as evidential data. such steganalysis attacks were developed to identify the In fact, various ways exist to store such data, including: a regular patterns used to store the data inside cover [16]– metadata stream [21]; using a frame’s subtitles (as available [18]. The works on Stego Color Cycle (SCC) approach in current surveillance systems) [22]; or through steganogra- [19], and Magic LSB technique [12] are recent additions to phy. However, a metadata stream is not used in this paper’s the field of steganography. The principal advantage of the method because storing data in different channels can cause LSB technique is the inherent simplicity of embedding and delays in transmission and, therefore, can introduce latency decoding process. Furthermore, the decoding of such stego- and misalignment in the retrieval module. Similarly, saving images is affected by the levels of noise, compression, quan- data in subtitles directly threatens the secrecy of data and, tization etc. in the communication channel. However, for the more importantly, disrupts the actual image content, which proposed work, actual transmission of the stego-image over may be important to data controllers when performing a a communication channel is not required. These facts form query-based search. Therefore steganography is selected as the basis of the choice of using LSB steganography in this a means of storage, as it avoids the delays of side-channels work. and the insecurity of subtitle storage [23].

264 Research has demonstrated that storing information in the least signiﬁcant bits of image pixels does not affect the PSNR value and, therefore, for some purposes is attractive. However, using regular patterns of storing information in image pixels do not promise improved security. Here we introduce random pixel selection from a pixel’s neighbourhood, and storing the path as steganoraphic data. Further to improve the random pixels selection the seed pixel is also selected using a mathematical function which means every IoT device can have its own unique implementation. The following algorithm describes the steganography method used. Moreover, this method produce different pattern for each frame of the video and therefore, is more robust and secure against attacks. For storage, two bits per color channel of a pixel are used for data (1 bit) and path (1 bit) storage, thereby allowing for 6 bits in a given pixel. For path selection we need three bits to generate eight different codes for eight neighbourhoods of a pixel as shown in Figure 3. Therefore, last bit of each color channel represents the path to the next pixel whereas the second last pixel stores the data bit as shown in Fig. 4. In this way each selected pixel will store 3 bits of the data, along with the code of the path to next pixel. It is to Figure 2: Life cycle from video capture, embedding evi- be noted that the distance between two randomly selected dential data and current frame hash, and edge storage to pixels is set at 5, and the codes shown in Fig. 4 only denote decryption, validation of evidential data and current frame the directions of the pixels to be selected, to ensure that hash, and onward video transmission no two pixels may have overlapping neighbourhood. Using this method 171 pixels will be required to store 512 bits of evidential data that consist of 256 bits of salted hash and 256 bits of frame’s own hash along with the path to be followed for the storage and retrieval of hashes. For seed pixel selection following mathematical function can be used: seedP ixel =(center of the image ± offset) (1) Here we chose an offest equal to +5 for odd numbered frames, −5 for even numbered frames. Therefore, for frames number 1, 3, 5, 7, ... the seed pixel is center of the image +5 and for 2, 4, 6, 8,... it is center of the image − 5. Figure 3: Codes for random selection of neighbourhood As illustrated in the overview of Fig. 2, a chain-of- pixels for path storage evidence (shown in Fig. 1) is stored along with the orig- inally captured visual data so that the authenticity of the R G B 1011010 1 101010 1 0 0110101 1 Figure 4: A typical pixel: Red bits are used to store data bits whereas blue bits will store path to the next selected pixel

video content can be established afterwards. The following discussion assumes that the system requires to store 512 bits of information in the cover media, out of which the ﬁrst 256 bits comprise of the salted hash (evidential data) and the next Figure 1: Creation of a chain-of-evidence, followed by 256 bits denote the frame’s own hash. To store the 512 bits, encryption and storage at the network edge 171 pixels are required since the proposed method stores 3 bits of data per pixel.

265 Figure 5: Random pattern of pixels selected to store chain-of-evidence using LSB Steganography in video imagery, results on two different videos are shown.

A detailed description of the steps in the method is given It is important to mention that for the ﬁrst two frames next, with the frames indexed by i: the salted hash is created by generating two AES keys 1) Creation of a 256 bit salted hash (S# = Fi−1#+ using a key management module. However, these keys do Fi−2#) from the hashes of the previous two frames, not need to be stored or sent to the receiver because they thus protecting against video frame insertion forgery. are only employed to generate a salted hash that is stored 2) Embedding of evidential data in randomly selected 85 in the initial two video frames for their own hash calculation. pixels in video frame. Additionally, zeros are embedded in the remaining randomly selected 86 pixels’ data The retrieval process works as follows: bits to act as a container for a hash of current frame. • Retrieve the private key (Kp) from the hardware wallet 3) Calculation of the hash (A#) for the current frame for the corresponding video session and then decrypt (Fi). the video. 4) Insertion of A# in Fi’s in reserved bits (from Step-2) • For each video frame, extract only the corresponding by replacing the zeros set previously. frame’s hash A# from the pixels storing that frame’s 5) Encryption of a session of video frames by means hash and place zeros in its place. of the Advanced Encryption Standard (AES) [24] and • Recalculate the hash of the frame, which now contains storage of the private key (Kp) in a hardware wallet the evidential (salted hash of previous frames) and along with the corresponding session id. image data.

266 • Match the calculated hash with the extracted hash. If ’SampleVideo 1280x720 1mb.mp4’1 and ’Atrium’ 2. The there is a match, the video is proven to be authentic first video is 1 MB in size with SVGA resolution while and, in the case of a mismatch, the video data have the Atrium video is 2.43 MB. For performance evaluation, been compromised and cannot be validated. processing time and video quality metrics were determined Figure 5 shows the highlighted pixels where the evidential and compared with classic LSB method and two recently data is stored in video frames and it can be seen that every published methods, SCC [19] and Magic LSB technique frame has a different pattern to store this data. Furthermore, [12]. SCC does the steganography in the RGB channels in a the data is stored uniformly in RGB color channels and uses cyclic fashion, whereas the Magic LSB uses a much complex only two bits per channel which makes it undetectable by method of converting an RGB image into an equivalent HSI the human eye and harmless to image content. It should be plane and then rotating the I component at four different noted that the proposed approach is specially suitable for angles. Each rotated I component of the image stores one the surveillance video use case where no image processing block of data (divided into four blocks) and then RGB image is allowed on the data captured, and all modifications of is reconstructed form HSI image. Although complex and the video data need to be identifiable for ascertaining the difficult to break, the scheme is too costly for IoT devices authenticity of the video. and therefore, rationalized the development of the proposed method. IV. ENCRYPTION A. Timings For encryption there exist a number of block-based, sym- The timings for the processing of a video stream across metric encryption techniques, such as the DES, Blowfish, the four main security steps has been calculated. The chain- RC5, and AES algorithms [24], all of which avoid the need of-evidence and steganographic processing consumes rel- for a Public Key Infrastructure in the case of asymmetric atively lesser time compared to the encryption/decryption cryptography. Even so, in terms of computational overhead processes i.e. 0.01 seconds and 0.02 seconds respectively their calculations still involve considerable processing, es- (for 100 frames). Overall, security processing at SVGA pecially if real-time operation on edge devices is required. resolution on the Raspberry Pi-4 supports a frame rate of However, considering the sensitivity of the surveillance data, 25 fps, which is suitable for surveillance videos. Notice that the industry-standard AES algorithm was selected. other processing in the video pipeline, such as compression, Here, to simplify the procedure, encryption keys are will result in a limited amount of start-up latency. refreshed after each video recording session, which is one hour by default, and a log file is also kept. Longer or short B. Steganographic conditions periods of refresh time can be set, depending on a camera’s location. With additional computational and networking Table I presents distinctive features of the proposed algo- costs, key security might be enhanced by any standardized rithm as compared to other methods. It can be seen that the key-management scheme [25]. proposed method focuses on hiding specific security related information and therefore, it is very small as compared to V. R ESULTS other secret messages and resulted in a very high PSNR value. The proposed algorithm has been compared with three The proposed security system has been developed for other algorithms for which the algorithm was either easy to resource constrained devices and, therefore, has been tested implement or was already available in public domain such on a Raspberry Pi-4 board with 4 GB RAM and a 64- as Classic LSB, SCC and Magic-LSB. The comparison has GB Secure Digital (SD) memory-card for data storage. been performed in light of the fact that a reliable technique The Pi was loaded with Python and OpenCV modules to for steganography needs to fulfil the following conditions process the videos, along with the Crypto cipher module [26]: for AES encryption. The AES block size was 64-bits, with • Invisibility: The data concealed in randomly selected AES operating as a stream cipher in Output Feedback pixels of each frame is not visible to the human eye (OFB) mode. OFB mode avoids the overhead of padding because it is stored in the LSB of only 171 pixels and, if transmission errors occur, these are not necessarily and a human eye cannot identify abnormalities in these propagated, allowing partial recovery of the stream. The key scattered pixels of a 640 × 480 pixels/frame or a size was 128 bits rather than 256 bits, in order to reduce the larger image. The Structural Similarity (SSIM) (output computational overhead. Compression was applied accord- range 0 to 1) [27] of the video with steganographic ing to the default MPEG-4 (MP4) settings for the board. data is very high, taking into account the response of Surveillance video data was captured by means of a Pi- the human visual system, as can be seen in Fig. 6c, camera V2, having a resolution of 3280×2464 pixels/frame and using the maximum frame rate of 30 frames/s (fps). 1https://vimeo.com/254653368 Also tested were publicly available evaluation videos, named 2https://www.jpjodoin.com/urbantracker/dataset.html

267 Table I: Features based comparative analysis of proposed system with some other steganographic methods Ref. Year Image/Video PSNR Encryption Hidden Data [9] 2011 image 62.1 Filter bank cipher Secret Message

[10] 2013 image 56.72 XOR Secret Message

[11] 2014 image 74.39 AES Secret Message

[12] 2016 image 62.67 Multilevel encryption (MLE) Secret Message

[14] 2017 image 63.00 RC4 and Pixel Shufﬂing Secret Message

[15] 2018 image 68.90 AES Secret Message

Proposed 2020 video Average PSNR of 100 video AES Chain of Evidence Algorithm frames= 80.47

implying that the original video and the video with thermore, the algorithm is proposed for a specific prob- data inserted are very similar. The plots show the effect lem of surveillance video data stored on the network of steganographic embedding of the evidential data edge and according to legal requirements. The video using four different methods namely Classic LSB, SCC, data are not allowed to leave their storage site unless Magic LSB and the proposed technique. In fact, the they are required for law & enforcement purposes and average SSIM remained as 0.99 using the proposed only if they are AES-encrypted. method, which confirms that the hidden evidential data • Steganographic quality: As previously mentioned, the are visually unidentifiable and the addition of the cur- presence of steganography is commonly detected by rent frame’s hash makes little difference to that finding. finding the changes in video quality [27], which is Furthermore, comparison of the proposed method with demonstrated in Figs. 6a and 6b. The average PSNR SCC, Magic LSB and Classic LSB also revealed that after steganography is around 80.0 dB, indicating high SSIM of the proposed method remained consistently similarity. This is attributed to the small amount of data higher for the whole video than any other method, embedded in the image (only 512 bits) and secondly, thereby ascertaining the superiority of the proposed only two bits per pixel color have been used for method over others in maintaining the invisibility of data hiding. Similarly, the average MSE between the the hidden data. original video frames and those with steganographic • Payload capacity: The payload of the proposed method data embedded is 0.01. This implies that it will be is 2 bits per color component per pixel and the data very difficult to detect the presence of steganography stored only uses 171 pixels of the whole image, which by means of objective video metrics, as was also shown is particularly small. Further, due to a random pattern for SSIM in Fig. 6c. The performance of Magic LSB of pixels storing the evidential data, it is completely method is also comparable to the proposed method with imperceptible to the human visual system. Although low MSE and high PSNR. the payload of Magic LSB is also small with the use of only one bit per pixel, but due to its complexity and processing requirements it is not very suitable for Table II: Prominent features of surveillance products and the resource constrained devices. proposed solution • Robustness against statistical attacks: Common meth- Features Hikvision Swann Smart Sony Intelligent Proposed Solution Smart Camera Video Analytics Camera Camera ods used to detect steganography are through taking the [28] [29] [30] Mean Squared Error (MSE), the SSIM index and the Object Detection Peak Signal-to-Noise Ratio (PSNR). For all of these Motion Detection video quality metrics, the original video frames are re- Face Detection quired as a reference. However, in the proposed method, Activity Detection X X X the captured videos are not stored without appending Chain-of-Evidence XX X security information and, therefore, the original data Storage Cloud Cloud Cloud Edge (videos) are always similar to the retrieved data, a Solution-by-Design. The proposed solution is also compared with some well- • Non-suspicious files: The file formats and file sizes known products in the marketplace, from where it can be remain the same during storage, as would be the case seen that the solution is competitive with these products, as without embedded information to avoid suspicion. Fur- is evident from Table II.

268 side the video content, with overall encryption. Though relatively simple, the method is not only a solution-by- design but is also capable of being implemented on resource constrained devices such as the Raspberry-Pi. Moreover, the proposed method can also be used in the context of videos shared on social media and through mobile phones/devices. The video data can be marked with unique identiﬁcation using proposed salted hashes at source that can later be used to verify the originality of the shared video content, establishing trust and authentication in a variety of settings like surveillance and social media.

(a) PSNR of Pi-camera captured video REFERENCES

[1] A. Brunetti, D. Buongiorno, G. F. Trotta, and V. Bevilacqua, “Computer vision and deep learning techniques for pedestrian detection and tracking: A survey,” Neurocomputing, vol. 300, pp. 17–33, 2018.

[2] R. D. Singh and N. Aggarwal, “Video content authentication techniques: A comprehensive survey,” Mult. Syst., vol. 24, no. 2, pp. 211–240, 2018.

[3] K. Sitara and B. M. Mehtre, “Digital video tampering detection: An overview of passive techniques,” Digital Investiga- tion, vol. 18, pp. 8–22, 2016. (b) MSE of Pi-camera captured video [4] D. A. Rodr´ıguez-Silva, L. Adkinson-Orellana, F. Gonz’lez- Castano, I. Armino-Franco, and D. Gonz’lez-Martinez, “Video surveillance based on cloud storage,” in IEEE 5th Int. Conf. on Cloud Comput., 2012, pp. 991–992.

[5] M. N. Asghar, N. Kanwal, B. Lee, M. Fleury, M. Herbst, and Y. Qiao, “Visual surveillance within the EU General Data Protection Regulation: A technology perspective,” IEEE Access, vol. 7, pp. 111 709–111 726, 2019.

[6] P. Voigt and A. Von dem Bussche, “The EU general data protection regulation (gdpr),” A Practical Guide, 1st Ed., Cham: Springer Int. Publ., 2017.

[7] R. Chandramouli and N. Memon, “Analysis of LSB based image steganography techniques,” in IEEE Int. Conf. on (c) SSIM of Pi-camera captured video Image Proc., vol. 3, 2001, pp. 1019–1022. Figure 6: Performance comparison of the proposed method with prior art. The comparison was performed for the [8] M. Sutaone and M. Khandare, “Image based steganography proposed method, SCC, Classic LSB, and the Magic LSB using lsb insertion,” in IET Conference on Wireless, Mobile and Multimedia Networks. Stevenage, UK: IET, 2008, pp. techniques. 146–151.

[9] S. M. Karim, M. S. Rahman, and M. I. Hossain, “A new VI. CONCLUSION approach for lsb based image steganography using secret key,” in 14th international conference on computer and information Provision of a chain-of-evidence in surveillance videos is technology (ICCIT 2011). IEEE, 2011, pp. 286–291. not only a computer security application but a legal requirement for current surveillance systems, due to the ratiﬁcation [10] S. Saraireh, “A secure data communication system using of the EU’s GDPR and privacy laws in other countries. cryptography and steganography,” International Journal of The proposed method provides a comprehensive solution by Computer Networks & Communications, vol. 5, no. 3, p. 125, storing such evidence steganographically embedded along- 2013.

269 [11] M. R. Islam, A. Siddiqa, M. P. Uddin, A. K. Mandal, [20] H. Gilbert and H. Handschuh, “Security analysis of SHA-256 and M. D. Hossain, “An efﬁcient ﬁltering based approach and sisters,” in ACM Symp. on Applied Comput., 2003, pp. improving lsb image steganography using status bit along 175–193. with aes cryptography,” in 2014 International Conference on Informatics, Electronics & Vision (ICIEV). IEEE, 2014, pp. [21] Y.-T. Tseng, P. Jurczyk, S. Watson, and M. Dalcin, “Merged 1–6. video streaming, authorization, and metadata requests,” Oct. 2 2018, US Patent 10,091,192. [12] K. Muhammad, M. Sajjad, I. Mehmood, S. Rho, and S. W. Baik, “A novel magic LSB substitution method (M-LSB-SM) [22] D. Stanescu, M. Stratulat, B. Ciubotaru, D. Chiciudean, using multi-level encryption and achromatic component of an R. Cioarga, and M. Micea, “Embedding data in video stream image,” Mult. Tools and Apps., vol. 75, no. 22, pp. 14 867– using steganography,” in 24th Int. Symp. on Applied Compu- 14 893, 2016. tational Intelligence and Informatics, 2007, pp. 241–244.

[13] W. S. Sari, E. H. Rachmawanto, C. A. Sari et al., “A good [23] I. Cox, M. Miller, J. Bloom, J. Fridrich, and T. Kalker, Digital performance OTP encryption image based on DCT-DWT watermarking and steganography. Amsterdam: Morgan steganography,” Telkomnika, vol. 15, no. 4, pp. 1987–1995, Kaufmann, 2007. 2017. [24] C. Paar and J. Pelzl, Understanding cryptography: A textbook [14] M. H. Abood, “An efﬁcient image cryptography using hash- for students and practitioners. Sci. & Business Media, 2009. lsb steganography with rc4 and pixel shufﬂing encryption algorithms,” in 2017 Annual Conference on New Trends [25] S. M. Bellovin and R. Housley, “Guidelines for cryptographic in Information & Communications Technology Applications key management,” in Symp. on Research in Security and (NTICT). IEEE, 2017, pp. 86–90. Privacy, 2005.

[15] A. Shifa, M. S. Afgan, M. N. Asghar, M. Fleury, I. Memon, [26] M. Douglas, K. Bailey, M. Leeney, and K. Curran, “An S. Abdullah, and N. Rasheed, “Joint crypto-stego scheme for overview of steganography techniques applied to the protec- enhanced image protection with nearest-centroid clustering,” tion of biometric data,” Mult. Tools and Appls., vol. 77, no. 13, IEEE Access, vol. 6, pp. 16 189–16 206, 2018. pp. 17 333–17 373, 2018.

[16] S. Rajendran and M. Doraipandian, “Chaotic map based ran- [27] Z. Kotevski and P. Mitrevski, “Experimental comparison of dom image steganography using LSB technique.” IJ Network PSNR and SSIM metrics for video quality estimation,” in Security, vol. 19, no. 4, pp. 593–598, 2017. International Conference on ICT Innovations, 2009, pp. 357– 366. [17] N. Patel and S. Meena, “LSB based image steganography using dynamic key cryptography,” in IEEE Int. Conf. on [28] “Smart camera, Hikvision Digital Technology Co.” Emerging Trends in Commun. Technol., 2016, pp. 1–5. https://www.hikvision.com/, accessed: 01-05-2020.

[18] X. Zhou, W. Gong, W. Fu, and L. Jin, “An improved method [29] “Smart video analytics camera, Swann Communications Pty. for LSB based color image steganography combined with Ltd.” https://www.swann.com/us/swwhd-intcam, accessed: cryptography,” in IEEE/ACIS 15th Int. Conf. on Computer 01-05-2020. and Info. Sci., 2016, pp. 1–4. [30] “Intelligent camera, Sony Europe B. V.” [19] K. Muhammad, J. Ahmad, and N. Rehman, “A secure cyclic https://pro.sony/en CL/products/ip-cameras/video-security- steganographic technique for color images using randomiza- g6-intelligent-surveillance, accessed: 01-05-2020. tion,” Tech. J. (Taxila), vol. 19, no. 3, pp. 57–64, 2014.

270