This paper presents a digital rights protection scheme for both colour and grayscale images using a novel approach that combines watermarking and cryptography. The schema involves two parties: the owner of the digital rights and a generic user who acquired some rights on a copy of the image that will be watermarked. The watermark, a QR code derived from a signed \License Agreement", is repeatedly inserted, and scrambled, by the image right's owner, into the frequency components of the image, thus producing the watermarked image. The schema, a non-blind type, achieves good perceptive quality and fair robustness using the 3rd level of the Discrete Wavelet Transform. The experimental results show that, inserting more occurrences of a scrambled QR code, the proposed algorithm is quite resistant to JPEG compression, rotation, cropping and salt & peeper noise.
data into the image, the original copy is decomposed into its 3rd level Discrete Wavelet components, where the watermark is inserted into the lower frequency components. We chose a QR Code as type of watermark [KLH13, CCC14] because of its error correction capability and because it is visually perceptible. In order to improve imperceptibility and extraction performances, we insert it into the host image more times, in a keyscrambled version. After watermark insertion we need to perform an Inverse Discrete Wavelet transform to obtain the watermarked image. For such a non-blind schema, to extract the watermark is necessary to provide the original image, while the QR Code is used only to compare its payload to the one extracted from the watermarked image. 2
Digital watermarking could be considered a sort of steganography. Image watermarking techniques can be classi ed [CCCM10, HH13, Rak13, HYAAQ12, KK11] into di erent categories according to the type of domain in which data embedding takes place and the type of information is needed to extract the watermark. There are mainly two domain types [RCSD13]: spatial and frequency. Regarding what is needed to extract the watermark from the image we can divide the cases into blind, semi-blind and non-blind systems [NVR16]. A blind watermark, or public watermarking algorithm, requires neither the cover image (original image) nor the embedded watermark to extract it from the watermarked image; a semi-blind, or semi-private scheme, requires only the watermark and nally a blind scheme requires at least the cover image. Spatial image watermarking techniques [RMK15] are commonly used in a \pure" steganographic context because, hiding data into the least signi cant bits of an image, achieves to embed large quantity of data but the watermark is not robust to common manipulations like JPEG compression [TCZA14]. In frequency domain the most common techniques [SK15] are based on Discrete Cosine Transform (DCT) or Discrete Wavelet Transform (DWT). Due to the block divide algorithm used generally by DCT watermarking, these techniques are robust against JPEG compression but are not resistant to rotation, translation and image cropping. Instead, embedding data using the DWT assures good robustness against the most popular image manipulations [XSSL04]. Wavelet transform is a modern technique that was rst used to study non-stationary events like earthquakes. Digital images properties can be better expressed through a wavelet transform since the frequency components are quickly varying around the image area. The main di erence from the DCT is that the DWT transform is based on a sum of scaled and shifted \mother" wavelets that have a limited duration. Through the wavelet decomposition the original signal can be represented by its coe cients which contains the spatial information. Each level of a DWT produces four types of coe cients: LL, or approximation coe cients, that represent the low frequency part of the image (most of information) and the details coe cients LH, HL and HH (vertical, horizontal and diagonal). Fig. 1 illustrates the sub-band decomposition of an image using 2D wavelet transform after 3 levels of decomposition. In every level the decomposition is obtained on the LL component of the previous level. The original signal can be completely reconstructed performing the Inverse Wavelet Transformation [Mal89] on these coe cients. In order to achieve a good visual imperceptibility, according to the spectral sensitivity of human eye, the blue component of a colour image is most suitable for hiding data [Gol09]. Data hiding system performances [Ber14] are described in terms of imperceptibility, embedding capacity and robustness. For digital watermarking the most important are imperceptibility and robustness.
To measure visual imperceptibility between two images the most used indices are Mean Squared Error (MSE), Peak Signal to Noise Ratio (PSNR) and Structural Similarity [WBSS04]:
MSE =
1 m n m n X X(X(i; j) i=1 j=1
X0(i; j))2 PSNR = 10 log10
MAXi2
MSE SSIM =
(2 i j + C1)(2 ij + C2) ( i2 + j2 + C1)( i2 + j2 + C2) where: m and n are the number of rows and columns of the image expressed in pixel, X(i; j) is the value of the pixel at row i and column j of the original image, X0(i; j) is the value of the pixel at row i and column j of the watermarked image;
MAXi is the biggest value of a pixel, MSE is the Mean Squared Error; ,
and ij are, respectively, mean, standard deviation and correlation, and C1, C2 are constants.
Digital image watermarking robustness can be evaluated in terms of correct watermark extraction after an image alteration. Common image manipulations regard JPEG compression, rotation, cropping and additive noise.
The Quick Response Code (QR Code) was developed during the 80's in Japan by Denso Wave company and used to monitor automotive spare parts and became a standard more than 20 years ago. It consists of black squares arranged into a white grid and uses Reed-Solomon error correction codes [ISO15]. According to its visual identi cation and error correction property it achieves good robustness against image degradation. 3
In this section we describe the procedure to both generate and verify the watermarked image, starting from the original image and its digital rights speci cation. Suppose Bob is the owner of the original image and Alice is interested in buying a copy of it, whereas Bob prefers to keep his digital rights. So they create a document containing at least these items: original image thumbnail and hash;
digital rights regarding the image;
This document is then signed by both parties using their own private keys. Fig. 2 presents the watermark injection procedure. Starting from the original image (Fig. 4(a)) and the signed document, Bob produces the watermarked image for Alice by the following steps: computes the approximation coe cients of level 3 (ACLL3) by performing a third-level decomposition of the image using a wavelet (blue component in case of colour image); produces a QR-code encoding the hmac-sha256 of the signed document using his private key; derives a scrambling key from a hmac-sha256 of a password and use it to scramble the QR-code repetitions necessary to t the size of the ACLL3 (N M ) (Fig. 3(a)) of the image obtaining WIM (N M ) (Fig. 3(b)); insert the watermark into the approximation coe cients of level 3 of the watermarked image WACLL3(i; j) = ACLL3(i; j) + k WIM(i; j), i = 1; 2; : : : ; N and j = 1; 2; : : : ; M , with k = 20000 for colour images and k = 15000 for black and white ones; obtain the watermarked image (Fig. 4(b)) by performing the inverse discrete wavelet transform. The hash of the resulting watermarked image is appended to the signed document and nally signed once again by both Bob and Alice. The watermark extraction procedure (Fig. 5) can be done only by Bob because we need his private key and the unscrambling password. Starting from the original image (Fig. 4(a)) and the watermarked image it is necessary to: compute the approximation coe cients of level 3 by performing a third-level decomposition of the image using a wavelet (blue component in case of colour image) for both original (ACLL3) and watermarked image (WACLL3); reconstruct WIM:
WIM(i; j) =
0 otherwise
t (a)
(b) for i = 1; 2; : : : ; N and j = 1; 2; : : : ; M , with t = 40 for colour images and t = 22:5 for black and white ones, obtaining a scrambled watermark (Fig. 3); descramble by using the key derived from the hmac-sha256 of Bob's password; compute hmac-sha256 of the document using Bob's private key; recover the QR-code from the single QR-code repetions occurring in the descrambled image (Fig. 3) and verify if the decoded value is equal to the hmac-sha256 of the document (for payload extraction is used either each single extracted repetition of a QR code either a reconstructed QR code based on majority pixel value matching, upon 1 to the maximum value of them). 4
We implemented this technique in MATLAB using as original images the ones represented in Fig. 7.1 We used the open-source library \libqrencode" [Ken18] for QR code generation and the \quirc" library [Dan12] for QR code decoding. All images are of equal size 512 512 and were tested with di erent Alice and Bob keys with the rst 20 wavelets of the Daubechies family [Dau92]. The results show that the algorithm achieves fairly good results in terms of imperceptibility (Fig. 8). For every image and wavelet the robustness of the watermarking schema was evaluated through the correct extraction and decoding of the QR code from the watermarked image that has 1Source code, available under the GPL-3.0 license, can be dowloaded from http://www.diag.uniroma1.it/~damore/ watermarking/src/sorgenti.tar.gz. been JPEG compressed (quality factors ranging in [50; 100]), =4 rotations (rotations maintaining the bounding box of the original image imply also cropping) and with presence of salt & pepper noise. Salt & pepper noise case extraction (SPNOW) was evaluated also preprocessing the watermarked image with a median lter (SP) [CT15]. The results in Fig. 9, 10, 11, 12 and 13 show that db7 wavelet obtains the best results among the rst 20 wavelets of Daubechies family and that salt & pepper noise impacts more than the other attacks on watermark robustness. Further tests were performed on the image of Fig. 14 by taking 4 occurrences of di erent size: 4641 3315, 1280 914, 1920 1371 and 640 457 pixels. We achieved a fully correct reconstruction in the rst 3 sizes, con rming the robustness of the proposed schema. On the other hand the results for the smallest sample (640 457) in Fig. 15 con rm how salt & pepper noise has a deeper impact a ecting the reconstruction, when one the size of the image is smaller than 512 pixel. This is explained by the reduced number of QR code occurrences, clearly insu cient to appropriately obtain the correct QR code (Fig. 16).
A novel digital rights protection scheme based on the 3rd level DWT image watermarking technique has been designed, implemented and evaluated. This schema can embed scrambled intellectual property references within an image with almost no e ect on its quality. Experimental results show that such a schema provides quite good quality and robustness. Further analysis could be done by investigating on single image adaptive constants k (see Section 3) in order to improve the robustness and the reconstruction of the QR code from its occurrences. Another interesting application could be testing such a schema on the single frames of a video. [Ber14]
Fariba Ghorbany Beram. E ective parameters of image steganography techniques. International Journal of Computer Applications Technology and Research, 3(6):361{363, 2014.
Ji-Hong Chen, Wen-Yuan Chen, and Chin-Hsing Chen. Identi cation recovery scheme using quick response (qr) code and watermarking technique. Applied Mathematics & Information Sciences, 8(2):585, 2014. [CCCM10] Abbas Cheddad, Joan Condell, Kevin Curran, and Paul McKevitt. Digital image steganography:
Survey and analysis of current methods. Signal Processing, 90(3):727{752, 2010.
Pratap Singh Chahar and Vandana Vikas Thakre. Performance comparison of various lters for removing salt & pepper and speckle noises. International Journal of Computer & Communication Engineering Research (IJCCER), 3(5), 2015.
Trans. Information Theory, 36(5):961{1005, 1990.
I. Daubechies. Ten Lectures on Wavelets. Society for Industrial and Applied Mathematics, 1992.
[HYAAQ12] Nagham Hamid, Abid Yahya, R. Badlishah Ahmad, and Osamah M. Al-Qershi. Image steganography techniques: an overview. International Journal of Computer Science and Security (IJCSS), 6(3):168{187, 2012. [ISO15] [Ken18] [KK11] [KLH13] [Mal89] [NVR16] [Rak13] [RCSD13] [RMK15] [RVH96]
Information technology - Automatic identi cation and data capture techniques - QR Code bar code
symbology speci cation. Standard, International Organization for Standardization, Geneva, CH,
Feb
Kentaro Fukuchi. libqrencode, 2018. https://fukuchi.org/works/qrencode.
Jagvinder Kaur and Sanjeev Kumar. Study and analysis of various image steganography techniques. IJCST, 2(3):2229{433, 2011.
M Kim, D Li, and S Hong. A robust and invisible digital watermarking algorithm based on multiple transform method for image contents. In Proceedings of the World Congress on Engineering and Computer Science, volume 1, 2013.
Stephane G. Mallat. A theory for multiresolution signal decomposition: the wavelet representation. IEEE transactions on pattern analysis and machine intelligence, 11(7):674{693, 1989.
Mohammad Ali Nematollahi, Chalee Vorakulpipat, and Hamurabi Gamboa Rosales. Digital Watermarking: Techniques and Trends. Springer Publishing Company, Incorporated, 1st edition, 2016. Suresh Gawande Rakhi. A review on steganography methods. International Journal of Advanced Research in Electrical, Electronics and Instrumentation Engineering, 2(10), 2013.
Ratnakirti Roy, Suvamoy Changder, Anirban Sarkar, and Narayan C Debnath. Evaluating image steganography techniques: Future research challenges. In Computing, Management and Telecommunications (ComManTel), 2013 International Conference on, pages 309{314. IEEE, 2013. [SK15] [TCZA14] [Wav] [WBSS04] [XSSL04]
Sudhanshi Sharma and Umesh Kumar. Review of transform domain techniques for image steganography. International Journal of Science and Research, 2(2):1, 2015.
Hai Tao, Li Chongmin, Jasni Mohamad Zain, and Ahmed N Abdalla. Robust image watermarking theories and techniques: a review. Journal of applied research and technology, 12(1):122{138, 2014. Denso Wave. Answer to your questions about the QR Code. http://www.qrcode.com/en/ [accessed 15-Jun-2018].
Zhou Wang, Alan C. Bovik, Hamid R. Sheikh, and Eero P. Simoncelli. Image quality assessment: from error visibility to structural similarity. IEEE Trans. Image Processing, 13(4):600{612, 2004. Jianyun Xu, Andrew H Sung, Peipei Shi, and Qingzhong Liu. Jpeg compression immune steganography using wavelet transform. In Information Technology: Coding and Computing, 2004. Proceedings. ITCC 2004. International Conference on, volume 2, pages 704{708. IEEE, 2004.