Reversible authentication scheme for demosaicked ... - Springer Link

3 downloads 0 Views 497KB Size Report
Cheng Han Lin, Li Juan Lin, and Bo Jun Qiu. College of Computer and Information Sciences,. Fujian Agriculture and Forestry University. Fuzhou 350002, China.
Reversible authentication scheme for demosaicked images without false detection Xiao-Long Liu, Chia-Chen Lin , Cheng Han Lin, Li Juan Lin, and Bo Jun Qiu College of Computer and Information Sciences, Fujian Agriculture and Forestry University Fuzhou 350002, China Department of Computer Science and Information Management, Providence University, Taichung, Taiwan [email protected], {mhlin3,s1012565,s1032809,s1032842}@pu.edu.tw

Abstract. The integrity of an image transmitted over the Internet is easily breached, and as such, research into image authentication has become increasingly urgent. In this paper, a new ecient fragile watermarkbased authentication scheme for demosaicked images is proposed. In the proposed scheme, each watermark bit is embedded into two rebuilt components of each demosaicked pixel by using a designed authentication table. The watermarked image can be recovered back to the original demosaicked image if necessary. The experimental results demonstrated that our method preserves very satisfactory image quality and tamper detection accuracy. Most important, there was no false detection in the proposed scheme. Keywords: Image demosaicking; Image authentication; Tamper detection; Fragile watermarking

1

Introduction

To maintain the full integrity of digital images transmitted over the Internet, there has been considerable research for several years into fragile watermarkbased image authentication. Fragile watermark-based image authentication techniques are also known as strict authentication [1] and are designed to make the embedded watermark easily corrupted once the watermarked image is attacked by any kind of manipulation. Instead of resulting in meaningless encrypted data when directly applying traditional cryptography [2] to an original image, a watermark embedding procedure just modifies certain features of the original image and the resulting object remains a meaningful image. Most of the time, the distortion cannot be perceived by the human visual system and therefore it is not easy for malicious attackers to suspect that a watermark is placed in an image. Moreover, if an image is suspected of being manipulated or modified, 

© Springer International Publishing AG 2017 J.-S. Pan et al. (eds.), Advances in Intelligent Information Hiding and Multimedia Signal Processing, Smart Innovation, Systems and Technologies 63, DOI 10.1007/978-3-319-50209-0_38

313

X.-L. Liu et al.

314

the authenticity and integrity of that suspicious image can be easily verified by watermarking-based image authentication techniques. Moreover, when the image is declared unauthentic, the tampered areas can be detected or even restored using the detection or recovery functions provided by the authentication techniques. Digital still cameras are widely used in contemporary daily life to capture digital images of the real world. In general, a digital camera needs three sensors with three color filters to capture the R, B and G component of each color pixel, respectively. However, the hardware cost of having three sensors in a digital camera is quite expensive. To reduce hardware costs, a single sensor with a grid of dierent color sensors, called a color filter array (CFA) [3] is often used. The commonly used pattern for a CFA is the Bayer pattern, where only one component, i.e., red, blue or green, of each color pixel is significantly sampled by the corresponding grid. Inevitably, the use of CFA results in missing the other two components in each color pixel. To reconstruct the missing color components in each color pixel, an interpolation process called image demosaicking [4] is performed on the CFA sampled data, as shown in Fig. 1. Various image demosaicking schemes [5, 6] with varying reconstruction performance have been proposed.

Fig. 1. Example of image demosaicking.

In order to protect the integrity of demosaicked images, a reversible fragile watermark-based image authentication scheme for image demosaicking was proposed by Hu et al. [7] in 2014. Although good tamper detection accuracy can be achieved by using this scheme, the visual quality of the watermarked image is unsatisfactory. The distortion also enlarges when increasing the length of the fragile watermark. Furthermore, the watermarked image cannot be recovered back to the original demosaicked image. To provide reversibility and improve tamper detection accuracy and visual qualities, two reversible image authentication schemes were proposed by Lin et al. [8] and Hu et al. [9] in 2015 and

Reversible authentication scheme ...

315

2016, respectively. However, both of their schemes also introduce false detection, where several unmodified pixels are detected as modified pixels. This paper proposes a new ecient fragile watermark-based authentication scheme for demosaicked images. In the proposed scheme, each watermark bit is embedded into the two rebuilt components of each demosaicked pixel by using a designed authentication table. If the watermarked image is tampered, the tampered areas of the image are accurately detected, otherwise, the watermarked image can be recovered back to the original CFA form when necessary. The proposed scheme was compared with some related schemes and showed outstanding performance.

2

THE PROPOSED SCHEME

This section presents a detailed exposition of the proposed reversible fragile watermark-based image authentication scheme. After obtaining a sampled CFA image, any potential image demosaicking process can be applied on it to acquire a demosaicked image. The proposed watermark embedding procedure works directly on the two rebuilt components of the demosaicked pixels. Since the original sample components are not modified during the watermark embedding procedure, the watermarked image can be recovered back to the original demosaicked image with CFA sampling and image demosaicking processes. If the watermarked demosaicked image needs authentication, the proposed tamper detection procedure can be directly implemented on the suspicious image. 2.1

Watermark Embedding Procedure

Assuming the demosacked image with H*W pixels is to be watermarked. A fragile watermark with total number of H*W watermark bits is first generated. The watermark is a sequence of random values from 0 to 24, to generate it, the pseudo random number generator (PRNG) with a predefined seed is used to generate H*W random values. Each random value rv is then converted in to the watermark bit wb by Eq. (1): wb = rv mod 25.

(1)

For watermark embedding, a 5*5 authentication table (ACT) with values 0∼24 is predesigned, note that any ACT with values 0∼24 is suitable in this scheme, an example of it is shown in Fig. 2. Assume CP denotes the original color pixel with RGB components in the demosaicked image, where only one component of each CP is sampled by CFA, we denote it as o, and p and q denote the reconstructed two components of demosaicked pixel. Each watermark bit wb will be embedded into (p, q) by using the designed authentication table ACT. To embed each watermark bit wb into (p, q), we first compute the coordinate values x and y by Eq. (2): x = p mod 5, y = q mod 5.

(2)

X.-L. Liu et al.

316

Fig. 2. Example of a 5*5 authentication table (ACT).

where x and y represent the x-coordinate and y-coordinate in authentication table ACT, respectively. In accordance with coordinate x and y, the authentication value av = ACT[x, y] can be found in the authentication table. If av equals wb, no change is made on (p, q). Otherwise, we need to adjust p and q into p’ and q’. To adjust p and q, the coordinate value x’ and y’ of wb in authentication table is found. Then the difference between x and x’, y and y’ is calculated by Eq. (3): dx = x − x, dy = y − y  .

(3)

In order to minimize the distortion of modified pixel, dx and dy are then adjusted by Eq. (4): dx = dx + 5, if dx ≤ −2 dx = dx − 5, if dx > 2 dy = dy + 5, if dy ≤ −2 dy = dy − 5, if dy > 2.

(4)

The results include adjusted values to overcome the overflow and underflow problem, p’ and q’ will be calculated by Eq. (5): p = dx + p, if 1 < p < 254 p = dx + p + 5, if p ≤ 1 p = dx + p − 5, if p ≥ 254 q  = dx + q, if 1 < q < 254 q  = dx + q + 5, if q ≤ 1

(5)

q  = dx + q − 5, if q ≥ 254. An example of the watermark embedding procedure is described in the following. Assume the watermark bit wb to be embedded is 23 and the authentication table ACT is shown in Fig. 1. Given the color pixel CP= (157 , 82 , 229), where 157 is the sampled component o of CFA, 82 and 229 are the rebuilt components

Reversible authentication scheme ...

317

p and q, respectively. By Eq. (2), x is calculated as 2 and y is calculated as 4. The authentication value av of (p, q) is ACT[x][y]=ACT[2][4]=14, which is not equal to wb. Since wb=23 is located in ACT[0][2], x’=0 and y’=2. According to Eq. (3) and Eq. (4), dx is calculated as -2, and dy is calculated as -2. Therefore, with Eq. (5), p = 82 and q = 229 is modified into p’ = 80 and q’ = 227, respectively. As a result, the original color pixel CP= (157 , 82 , 229) will be changed into (157 , 80 , 227). 2.2

Tamper Detection Procedure

The goal of the tamper detection procedure is to detect whether the given image is modified or not, and locate the tampered area if the given image is detected as modified. To detect the H*W sized image, the original fragile watermark with total number of H*W watermark bit wb is first generated by PRNG the same as watermark embedding procedure. Then the extracted watermark with a sequence of H*W authentication values av is generated from the given image by Eq. (2) and looking up the authentication table. After that, we can determine whether each pixel is tampered or not by comparing wb and av. If the watermark bit wb is equal to extracted authentication value av, this pixel is regarded as clear pixels and marked as white. Otherwise, this pixel is regarded as modified pixels and marked as black.

3

Experimental Results and Discussions

Fig. 3. Eight test images.

This section presents some performance results of the proposed scheme. Subsequently, a performance comparison is done with other related work to demonstrate the superiority of the proposed scheme. All experiments were performed with eight commonly used full color images: Airplane, Baboon, House, Lena,

X.-L. Liu et al.

318

Pepper, Sailboat, Splash, and Tiffany, of the same size 512*512, as shown in Fig. 3. During experiments, the CFA images are captured by applying Bayer pattern to sample the original color images. Bilinear interpolation technique is selected to reconstruct the CFA image into demosaicked image. Table 1 presents the results for peak signal-to-noise ratio (PSNR) values of the demosaicked images reconstructed by bilinear interpolation and the watermarked images of the proposed scheme. The average visual qualities of the demosaicked images and the watermarked image were 29.86 dB and 29.83 dB, respectively. The PSNR values of image Baboon were significantly less than the average value, because it is a complex image, and its adjoining pixels have low correlation. Therefore, the image demosaicking process for Baboon has a significant distortion compared to the other images. However, the characteristics of the images do not influence the proposed watermarking scheme. The PSNR loss from the demosaicked images to the watermarked image was less than 0.1 dB, which indicates that the watermark embedding procedure of the proposed scheme incurs only a very tiny distortion to the demosaicked images.

Table 1. Results of PSNR Values Images Bilinear Interpolation Proposed scheme Airplane 31.24 31.12 Baboon 23.60 23.58 House 29.33 29.26 Lena 31.98 31.84 Pepper 30.12 29.99 Sailboat 28.36 28.29 Splash 35.39 35.00 Tiffany 29.83 29.56 Average 29.98 29.83

To demonstrate the tamper detection performance, some of the pixels in the water-marked images were first modified, and then detected by the proposed scheme. The tamper detection results of the proposed scheme are listed in Table 2, where NMP represents the total number of modified pixels in the tampered image, NDP is the number of detected pixels, NT DP is the number of true detected pixels, NF DP is the number of false detected pixels, PT D is the percentage of true detection in modified pixels and PF D is the percentage of false detection in modified pixels. The average percentage of true detection PT D of the proposed scheme is nearly 96%, this is because the extracted authentication value av of a modified pixel may be the same as watermark bit wb. The probability is 1/25 according to the designed authentication table. Therefore, only 4% of the modified pixels may miss by the detection. Most important, no false detection is introduced by the proposed scheme during tamper detection.

Reversible authentication scheme ...

319

Table 2. Detection Results of The Proposed Scheme Images Airplane Baboon House Lena Pepper Sailboat Splash Tiffany Average

NM P 29106 29106 29106 29106 29106 29106 29106 29106 29106

NDP 27931 27992 27936 27974 27934 27902 27923 27894 27935

NT DP 27931 27992 27936 27974 27934 27902 27923 27894 27935

PT D NF DP 95.96% 0 96.17% 0 95.98% 0 96.11% 0 95.97% 0 95.86% 0 95.94% 0 95.84% 0 95.98% 0

PF D 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00%

To further demonstrate the watermarking performance, we compared the proposed scheme with LSB substitution[10], Lin et al.’s scheme[8] and Hu et al.s scheme [9] by using the averaging results of the eight test images. Table 3 shows the performance results in PSNR, true detection rate PT D and false detection rate PF D of each scheme. The proposed scheme and scheme [8] achieve the best PSNR values (29.83dB), followed by LSB scheme (29.81dB) and scheme[9] (29.72dB). The true detection rate PT D of scheme [8], [9] and the proposed scheme are larger than 95%, which indicates that all of the schemes except LSB scheme (87.61%) can almost accurately detect every modified pixels in the tampered image. With respect to the false detection rate PF D , the proposed scheme and LSB scheme achieve the best performance, where no false detection is introduced. However, the PF D of scheme [8] and [9] are 1.21% and 7.10%, which means several clear pixels would be detected as modified pixels in their schemes. Thus, after considering the global performance of different schemes, the proposed image authentication scheme is demonstrably superior. Table 3. Comparison of The Related Schemes Schemes LSB[10] Lin et al.s[8] Hu et al.s[9] Proposed

4

PSNR 29.81 29.83 29.72 29.83

PTD 87.61% 98.20% 99.83% 95.98%

PFD 0.00% 1.21% 7.10% 0.00%

Conclusions

A reversible fragile watermark-based authentication scheme for demosaicked images was proposed in this paper. By utilizing a pre-designed 5*5 authentication

320

X.-L. Liu et al.

table, a watermark bit was embedded into the two rebuilt components of each demosaicked pixel. The tamper detection accurate rate was nearly 96% and there was no false detection. Comparing the proposed scheme with LSB substitution[10], Lin et al.’s scheme[8] and Hu et al.s scheme [9], the experimental results demonstrated that our method preserved the best image quality while maintaining satisfactory tamper detection accuracy. Most importantly, the proposed scheme provided demosaicked image reversibility without any false detection. Acknowledgments. This work is supported by MOST of Taiwan, No. 1032632-E-126-001-MY3.

References 1. Haouzia, A. Noumeir, R.: Methods for image authentication: a survey. Multimedia Tools and Applications. 39, 1, 1–46 (2008) 2. Rivest, R. L.: The MD5 message-digest algorithm. Internet Activities Board, Internet Privacy Task Force. (1992) 3. Lukac, R. Plataniotis, K. N.: Color filter arrays: Design and performance analysis. IEEE Transactions on Consumer Electronics. 51, 4, 1260–1267 (2005) 4. Menon, D. Calvagno, G.: Color image demosaicking: an overview. Signal Processing Image. 26, 8, 518–533 (2011) 5. Lu, Y.M. Karzand, M. Vetterli, M.: Demosaicking by alternating projections: theory and fastone-step implementation. IEEE Transactions on Image Processing. 19, 8, 2085–2098 (2010) 6. Ferrandas, S. Bertalmio, Caselles, M. V.: Geometry-based demosaicking. IEEE Transactions on Image Processing. 19, 3, 665–670 (2009). 7. Hu, Y.C. Chen, W.L. Hung, C.H.: A novel image authentication technique for color image demosaicking. in Proceedings of 2014 International Conference on Information Technology and Management Engineering. Hong Kong. (2014) 8. Lin, C.C. Lin, C.H. Liu, X.L. Yuan, S.M.: Fragile watermarking-based authentication scheme for demosaicked images. 2015 International Conference on Intelligent Information Hiding and Multimedia Signal Processing (IIH-MSP). Adelaide, SA. (2015) 9. Hu, Y.C. Lo, C.C. Chen, W.L.: Probability-based reversible image authentication scheme for image demosaicking. Future Generation Computer Systems. 62, 92– 103(2016) 10. Wang, R. Z. Lin, C. F. Lin, J. C.: ”Image hiding by optimal LSB substitution and genetic algorithm. Pattern recognition. 34, 3, 671–683 (2001)