- Research Article
- Open Access
Reversible Watermarking Using Statistical Information
© Ahmad Mahmoudi Aznaveh et al. 2010
- Received: 1 December 2009
- Accepted: 9 March 2010
- Published: 26 April 2010
In most reversible watermarking methods, a compressed location map is exploited in order to ensure reversibility. Besides, in some methods, a header containing critical information is appended to the payload for the extraction and recovery process. Such schemes have a highly fragile nature; that is, changing a single bit in watermarked data may prohibit recovery of the original host as well as the embedded watermark. In this paper, we propose a new scheme in which utilizing a compressed location map is completely removed. In addition, the amount of auxiliary data is decreased by employing the adjacent pixels information. Therefore, in addition to quality improvement, independent authentication of different regions of a watermarked image is possible.
- Watermark Image
- Adjacent Pixel
- Auxiliary Data
- Reversible Data
- Embed Watermark
Reversible watermarking, also called lossless data hiding, embeds the watermark data into a digital image in a reversible manner, that is, one can restore the original image without any degradation. Many techniques, focusing on capacity-distortion improvement, have been proposed during the last decade. In most of the previous work, channel degradation is not allowed; as a result, such schemes are highly fragile. This limits the usability of reversible watermarking only in lossless environments.
Lossless data embedding can be classified into the following categories: the first one utilizes additive spread spectrum [1, 2]; the second category compresses the selected image features for creating vacancy [3, 4], and employs this spare space for embedding; the third group, namely expansion-based methods [5–16], embeds the watermark data in some features by expanding them; these features are created by some decorrelation operator, and finally, some methods are based on histogram modification [17, 18] in which peak/zero points of the histogram either in spatial domain or transform domain are utilized for embedding.
Most of the existing reversible watermarking algorithms have a highly fragile nature in the sense that changing a single bit in the watermarked media may prevent the hidden data to be extracted. In addition, the restoration process would fail as well. This restricts the employing of reversible data hiding just to the cases in which there is a complete control over the watermarked data. Therefore, due to emerging applications of reversible watermarking schemes [19, 20], it makes sense to extend its scope to lossy environment.
The rest of this paper organized as follows: Section 2 reviews the concept of robust reversible watermarking. In the next section, our proposed method is introduced. The experimental results are presented in Section 4, and conclusion is drawn in Section 5.
Most of reversible watermarking methods, presented so far, have a highly fragile nature; nevertheless, there are some methods which can be deemed as semifragile techniques. In , Kalker and Willems provided a theoretical analysis of reversible data hiding techniques in which robust reversible watermarking is interpreted in three ways: firstly, it can refer to robustness of the embedded watermark. Next, it can refer to reversibility of the host signal, and finally, both payload and reversibility are considered. They focused on the third option and concluded that a robust reversible data hiding scheme exploits the side information available from the received data and also error correcting codes.
De Vleecshouwer et al. proposed a modulo-256 addition based on the classical patchwork algorithm in order to achieve a reversible watermarking scheme . Firstly, they map the histogram of each zone to a circle; then instead of the concept of average value, the position of the histogram on the circle is used as the discriminating factor. As a result, unreliable retrieval caused by wrapped around pixels impacting the average value is avoided. Due to embedding in nonoverlapping blocks and the small size of the auxiliary data, the recovery of a fragment of the payload using a grid alignment is possible for the cropped images. In addition, this method along with the message repetition is robust against JPEG compression just for extracting the payload. This scheme, however, is not free from salt and pepper noise. In addition, due to block-based embedding, the capacity is very low.
Ni et al.  proposed another semifragile reversible data hiding scheme based on the patchwork algorithm. They classified each block into four different categories and use different embedding schemes in order to avoid overflow and underflow. Their method employs error correction codes (BCH) in order to overcome some ambiguity raised during embedding and also to provide robustness against JPEG compression.
In order to enlarge the scope of reversible watermarking, a joint marking procedure is proposed in which a robust lossy watermark is first embeded and a reversible watermarking scheme is employed in the next stage . The information needed to invert back the marked media should be stored during the reversible embedding. In case of no attack, the original image can be recovered. It should be noted that due to high embedding capacity which is required in this method, the achieved quality is not acceptable.
Among different reversible watermarking schemes, the expansion based methods received more attention because they have the highest embedding capacity along with the lowest quality degradations. However, a location map is needed to determine the positions of the expanded values. This location map should be compressed in order to decrease its influence on embedding capacity. As a result, lots of efforts have been done to decrease the size of location map. Using location map, however, has some other shortcomings, for example, a single bit modification may break the entropy decoder synchronization. Furthermore, it is not possible to employ such embedding scheme in block-based manner.
As a result, it is necessary to remove the location map in order to prepare a less fragile method. There are some methods which do not use a location map. In , the locations of expanded values are determined based on some statistical analysis. However, it needs to transfer some information to the recipient separately. As a result, it cannot be consider as a blind scheme.
Coltuc and Chassery proposed a method in which using the location map is removed too . Robustness against cropping can be achieved by distributing and storing the auxiliary data close to the corresponding pixel pairs. Still, the size of auxiliary data especially in case of using distortion control is a major problem. In addition, in order to prevent some decoding ambiguity, the decoder should be informed about the error threshold in advance. They generalized their method in  in which a high capacity in a single pass of embedding can be achieved. However, the watermarked image quality is not satisfactory: the degradation is more annoying in edges. Moreover, in case of capacity control, the amount of auxiliary data which are used to specify the noncontainer elements increases intensively. In , an improved version of  is presented. However, due to dependencies in decoding process, it cannot be deemed robust against cropping. In addition, these dependencies may prevent the extraction of the watermark and recovery of the original image. In , this problem is solved through considering an extra state in the embedding phase. Still, this method cannot be robust against cropping.
In robustness point of view, introduced in , most of the presented semifragile reversible watermarking algorithms can be categorized in the first group in which just the extraction of the payload is possible in case of lossy operation. Although, the proposed methods in [11, 12] can recover the original image after cropping, the large number of auxiliary data is the main disadvantage of these methods.
In , we proposed an alternative transform to increase the quality of the marked image by employing the checkerboard structure. In , we expand the algorithm by utilizing a new error control strategy to decrease the size of auxiliary data. Since, this kind of reversible embedding algorithm does not employ a location map and the size of auxiliary data is negligible, it can be used to make the method more robust against some geometrical attacks.
Toward overcoming the highly fragile nature of reversible algorithms, two possible solutions are considered: the first one uses a block-based embedding; therefore, in case of some attacks such as cropping, it is possible to extract the embedded watermark using a grid alignment. In this case, the proper grid alignment should be recovered through an exhaustive search ; in addition, a low embedding capacity is another disadvantage of this group of algorithms. The other way is to design a reversible method in which the dependency to a location map or any auxiliary data become restricted. For example, by using the statistical information which can be retrieved from watermarked media; it is possible to determine the location of container elements. It should be noted that, this statistical information should be the same in both original and the watermarked image. This self-dependent reversible watermarking can be used as a way to decrease the fragile nature of the reversible methods.
Furthermore, and indicate the contribution factors of corresponding neighbors. By increasing the contribution factors, it is possible to embed more at the expense of higher degradation. Due to high spatial correlation in natural images, it is expected that differences between P and its neighborhoods will be small (which is the basic assumption in different expansion-based methods).
The sum of adjacent pixels should be remained intact so that the watermark extraction and recovery of the original host become possible; consequently, for embedding just half of the pixels, constructing a checkerboard structure is employed.
In (2), and w are congruent modulo M. As a result, it can be concluded from (2) that M symbols can be embedded in each pixel in a reversible manner. Therefore, the raw capacity is bpp. To achieve more capacity, it is possible to repeat the embedding procedure by changing the role of two pixel groups.
In this case, it is necessary to distinguish between transformable and nontransformable pixels. Therefore, similar to , one symbol is assigned to decide between the container elements and noncontainers. Since one can embed M symbols with keeping the ability of recovering the original value, we reserve the symbol `0' for noncontainer elements; hence, the watermark symbol can be chosen from . On the other hand, the range of auxiliary data may fall outside of this range since in case of overflow it is necessary to use negative corrective data. Thus, a prefix is utilized for encoding the range of with symbols.
Since the transform in (1) employs the image structure more fittingly, it can better preserve the image quality especially in edges, in comparison to similar methods [11–13]; this is because the human visual system is adapted to extract the structural information of a viewing scene .
where E is the number of embeddable pixels and N represents the number of pixel pairs.
In this case, missing the auxiliary data due to a synchronization attack prevents the algorithm to extract the payload and reverts back the original media. Thus, increasing the amount of auxiliary data will increase the fragility of the reversible watermarking methods.
The amount of auxiliary data especially after applying distortion control is increased dramatically which deteriorate the situation. As a result, to present a less fragile reversible method, reducing the number of auxiliary data is unavoidable.
Consequently, by using the information from the intact pixels, the amount of auxiliary data is decreased significantly. In other words, it is possible to embed the watermark in some elements independent from others. In this case, this scheme has the potential to be robust against synchronization attacks because the extracting and recovery for a fragment of a picture can be done independently.
In this section, firstly, the performance of our method is evaluated in terms of capacity and distortion. Then, the results in case of some lossy environment are explored.
For a fair comparison, the same expansion amount is utilized. For the proposed method and , 1.5 is considered as the contribution factors while in [11, 23] the simulation is performed for . In this case, the expansion amounts are similar in all cases. The experiments are conveyed on benchmarks of size .
As it is indicated, the proposed method outperforms the other methods especially in low embedding bit rate due to decreasing the required auxiliary data. In this case, just a less portion of the payload is occupied by the auxiliary data.
As it is depicted in above figures, decreasing the size of auxiliary data could restrict the introduced distortion. On the other hand, it decreases the dependency of the embedded watermark to the auxiliary data as well.
The number of auxiliary data in the proposed method.
A location map-free reversible watermarking scheme is proposed. Since the information of adjacent pixels is utilized, the size of the auxiliary data decreases. Therefore, the quality of the proposed method is improved significantly especially in low embedding rate. Furthermore, the negligible amount of the auxiliary data provides robustness against some geometric attacks such as cropping in which in addition to extracting the embedded watermark, the original image can be recovered.
The authors would like to thanks ITRC (Iran Telecommunication Research Center) for partially supporting this research.
- Macq B: Lossless multiresolution transform for image authenticating watermarking. Proceedings of the 10th European Signal Processing Conference (EUSIPCO '00), September 2000, Tampere, FinlandGoogle Scholar
- De Vleeschouwer C, Delaigle JF, Macq B: Circular interpretation of bijective transformations in lossless watermarking for media asset management. IEEE Transactions on Multimedia 2003, 5(1):97-105. 10.1109/TMM.2003.809729View ArticleGoogle Scholar
- Fridrich J, Goljan M, Du RUI: Lossless data embedding for all image formats. Security and Watermarking of Multimedia Contents IV, February 2002, San Jose, Calif, USA, Proceedings of SPIE 572-583.View ArticleGoogle Scholar
- Celik MU, Sharma G, Tekalp AM, Saber ELI: Lossless generalized-LSB data embedding. IEEE Transactions on Image Processing 2005, 14(2):253-266.View ArticleGoogle Scholar
- Tian JUN: Reversible data embedding using a difference expansion. IEEE Transactions on Circuits and Systems for Video Technology 2003, 13(8):890-896. 10.1109/TCSVT.2003.815962View ArticleGoogle Scholar
- Kamstra L, Heijmans HJAM: Reversible data embedding into images using wavelet techniques and sorting. IEEE Transactions on Image Processing 2005, 14(12):2082-2090.MathSciNetView ArticleGoogle Scholar
- Thodi DM, Rodríguez JJ: Expansion embedding techniques for reversible watermarking. IEEE Transactions on Image Processing 2007, 16(3):721-730.MathSciNetView ArticleGoogle Scholar
- Lee S, Yoo CD, Kalker TON: Reversible image watermarking based on integer-to-integer wavelet transform. IEEE Transactions on Information Forensics and Security 2007, 2(3):321-330.View ArticleGoogle Scholar
- Hu Y, Lee HK, Li J: DE-based reversible data hiding with improved overflow location map. IEEE Transactions on Circuits and Systems for Video Technology 2009, 19(2):250-260.View ArticleGoogle Scholar
- Sachnev V, Kim HJ, Nam J, Suresh S, Shi YQ: Reversible watermarking algorithm using sorting and prediction. IEEE Transactions on Circuits and Systems for Video Technology 2009, 19(7):989-999.View ArticleGoogle Scholar
- Coltuc D, Chassery J-M: High capacity reversible watermarking. Proceedings of the IEEE International Conference on Image Processing (ICIP '06), October 2006, Atlanta, Ga, USA 2565-2568.Google Scholar
- Coltuc D, Chassery JM: Very fast watermarking by reversible contrast mapping. IEEE Signal Processing Letters 2007, 14(4):255-258.View ArticleGoogle Scholar
- Coltuc D: Improved capacity reversible watermarking. Proceedings of the International Conference on Image Processing (ICIP '07), October 2007 3: 249-252.Google Scholar
- Chaumont M, Puech W: A high capacity reversible watermarking scheme. Visual Communications and Image Processing, February 2009, San Jose, Calif, USA, Proceedings of SPIE 72571H-9Google Scholar
- Aznaveh AMahmoudi, Mansouri A, Torkamani-Azar F: A new approach in reversible watermarking. Proceedings of the 8th International Workshop on Digital Watermarking (IWDW '09), 2009, Guilford, UK, Lecture Notes in Computer Science 5703: 241-251.MATHGoogle Scholar
- Aznaveh AMahmoudi, Torkamani-Azar F, Mansouri A: Toward quality improvement in location map free reversible watermarking. Proceedings of the 10th Pacific Rim Conference on Multimedia (PCM '09), December 2009, Bangkok, Thailand, Lecture Notes in Computer Science 5879: 867-876.Google Scholar
- Ni Z, Shi YUNQ, Ansari N, Su WEI: Reversible data hiding. IEEE Transactions on Circuits and Systems for Video Technology 2006, 16(3):354-362.View ArticleGoogle Scholar
- Tai W-L, Yeh C-M, Chang C-C: Reversible data hiding based on histogram modification of pixel differences. IEEE Transactions on Circuits and Systems for Video Technology 2009, 19(6):906-910.View ArticleGoogle Scholar
- Zhang X, Wang S: Fragile watermarking with error-free restoration capability. IEEE Transactions on Multimedia 2008, 10(8):1490-1499.View ArticleGoogle Scholar
- Coltuc D: On stereo embedding by reversible watermarking. Proceedings of the International Symposium on Signals, Circuits and Systems (ISSCS '07), August 2007 1-4.Google Scholar
- Kalker TON, Willems FMJ: Capacity bounds and constructions for reversible data-hiding. Security and Watermarking of Multimedia Contents V, February 2003, Santa Clara, Calif, USA, Proceedings of SPIE 604-611.View ArticleGoogle Scholar
- Ni Z, Shi YUNQ, Ansari N, Su WEI, Sun Q, Lin X: Robust lossless image data hiding designed for semi-fragile image authentication. IEEE Transactions on Circuits and Systems for Video Technology 2008, 18(4):497-509.View ArticleGoogle Scholar
- Coltuc D, Chassery JM: Distortion-free robust watermarking: a case study. Proceedings of the Security, Steganography, and Watermarking of Multimedia Contents IX, March 2007, San Jose, Calif, USA 65051N-8Google Scholar
- Jin HL, Fujiyoshi M, Kiya H: Lossless data hiding in the spatial domain for high quality images. IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences 2007, E90-A(4):771-777. 10.1093/ietfec/e90-a.4.771View ArticleGoogle Scholar
- Wang Z, Bovik AC, Sheikh HR, Simoncelli EP: Image quality assessment: from error visibility to structural similarity. IEEE Transactions on Image Processing 2004, 13(4):600-612. 10.1109/TIP.2003.819861View ArticleGoogle Scholar