Optimized Copyright Protection Systems with Genetic-Based Robust Watermarking

Optimized Copyright Protection Systems with

Genetic-Based Robust Watermarking

A_{National University of Kaohsiung, Taiwan, R.O.C.}

B_{National Kaohsiung University of Applied Sciences, Taiwan, R.O.C.}

Abstract

Research of robust watermarking algorithms is one of the major branches in digital rights management (DRM) systems. It is generally agreed that the quality of watermarked contents, the survivability of extracted watermark, and the number of bits embedded, need to be considered to evaluate how well the algorithm and hence the system is. However, these three metrics conflict with each other. Some previous works aimed at finding the tradeoff between quality and survivability by fixing embedded bits a constant. Besides quality and survivability, we take the number of embedded bits into account. With genetic algorithm, we designed an improved watermarking system that will obtain the reasonable quality, acceptable survivability, and practical capacity. Simulation results present the practical implementation of the proposed algorithm.

1. Introduction

Multimedia contents are easily spread over the Internet. Due to the ease of delivery and modification of digital files, the copyrights might be infringed upon. To deal with this, digital rights management (DRM) systems can prevent users from using such contents illegally [1]. In DRM systems, encryption and robust watermarking are two major schemes for applications [2][3]. The encrypted digital contents look like random patterns, which will cause the eavesdroppers to suspect the existence of hidden secrets. If one bit is received erroneously during transmission, part or whole of received data would not be decrypted, leading to the uselessness of such contents. For watermarking, the output contents and original counterparts look similar, or even identical from subjective point of view. During transmission, if some parts are received in error, they can partly be recognized and the copyright can be preserved. Thus, we focus on watermarking and its application in DRM in this paper.

In this paper, we use the digital images to represent the multimedia contents. It is generally agreed that for one watermarking algorithm, the output image quality (or

imperceptibility), the survivability, represented by the correct rate of extracted watermark (or robustness), and the number of bits embedded (or capacity) are the three most important factors to assess how good the algorithm is. Some trade off must be searched for because the three factors conflict with each other; hence, we employ genetic algorithm (GA) to find an optimized solution that can reach better imperceptibility, more robustness, and reasonable capacity. It can be directly applicable to DRM systems. This paper is organized as follows. In Section 2 we point out the need for optimization in a watermarking system. In Section 3 we then describe the proposed algorithm by modifying and extending previous works. Simulation results are demonstrated in Section 4. Finally, we conclude this paper in Section 5.

2. The need of optimization for robust

watermarking algorithm

2.1 Requirements for robust watermarking

As we stated in Sec. 1, the three major requirements for robust watermarking are imperceptibility, robustness, and capacity. Their interrelationships can be discussed as follows. And we can see why they conflict with each other.

¾ Watermark imperceptibility refers to whether the viewer can perceive the existence of embedded watermark or not. To make the watermarked image imperceptible, the watermark should be hidden into less significant parts, such as the least significant bits in the spatial domain or the high frequency components in the transform domain.

¾ Watermark robustness means the capability that the watermarked media can withstand intentional or unintentional media processing, called attacks, including filtering, resizing, or rotation. There are also benchmarks to perform attacks. To make the algorithm robust, the watermark needs to be hidden into more important parts, such as the most significant bits or the low frequency components.

iteration is the optimized watermarked image. Corresponding secret key with this image is also delivered to the receiver.

4. Experimental Results

We choose the test image Lena with the picture size of 512u512. The binary watermark with the size of 128u128 is prepared, shown in Figure 2(a). And we will compare with the results shown in [6]. Because in [6], authors used normalized correlation (NC) to represent the watermark robustness, and we use BCR here, hence we show extracted watermarks for subjective evaluation. Both [6] and our paper denote imperceptibility by using PSNR, we use Table 1 to make comparisons.

(a) watermark (b) BCR=0.8692 (c) BCR=0.8690

(d) BCR=0.9216 (e) BCR=0.9680

Figure 2. (a) Embedded binary

watermark with size 128u128. (b)-(e) Extracted watermarks with capacities 4, 3, 2, and 1 bit/block, respectively.

We choose the JPEG attack with quality factor QF=80 to validate the proposed algorithm. First, we compare the extracted watermarks in Figure 2. Figure 2(b) shows the one with capacity of 4 bit/block. Figure 2(c)-(e) illustrate those with capacity of 3, 2, 1 bit/block, respectively, leading to the watermark size of 128u96, 128u64, and 128u32. We can see that in Figure 2(c), only the upper three quarters can be recognized. The bottom quarter is intentionally set to bit 0 for comparison. Figure 2(d) and (e) also have similar phenomena. For embedding 3 or 4 bit/block, similar BCR values can be obtained. When decreasing the capacity to 2 or 1 bit/block, the BCR values grow. However, decreasing the capacity leads to be less meaningful in the extracted watermarks.

Table 1 makes comparison between the scheme in this paper and that in [6]. We can see that under the same capacity, we obtain better PSNR values, which implies the better imperceptibility. When we lower the embedded capacity, the PSNR values get higher. This is because less DCT coefficients get modified.

Table 1. Comparisons of capacity (in bit/block) and imperceptibility, represented by PSNR (in dB), between our algorithm and existing one.

Scheme Capacity Imperceptibility Existing ([6]) 4 bit/block 34.79 dB 4 bit/block 35.72 dB 3 bit/block 36.82 dB 2 bit/block 38.11 dB Proposed 1 bit/block 40.71 dB From the data in Figure 2 and Table 1 above, we can find out the improvements with the proposed algorithm. By taking the watermark capacity into account, our results can reach better imperceptibility and robustness.

5. Conclusion

In this paper, we discussed about the optimization of robust watermarking with genetic algorithms. By finding tradeoffs among robustness, capacity, and imperceptibility, we design a practical fitness function for optimization. Simulation results depict the improvements of our algorithm, hence the implementation of copyright protection system, and it is directly extendable to cope with a variety of attacks in the benchmarks. Other optimization techniques with different watermarking requirements will also be considered in the future.

6. References

[1] J. S. Pan, H.-C. Huang, and L. C. Jain (editors),

Intelligent Watermarking Techniques, World Scientific Publishing Company, Singapore, Feb. 2004.

[2] J. S. Pan, H.-C. Huang, L. C. Jain, and W. C. Fang (editors), Intelligent Multimedia Data Hiding, Springer, Berlin-Heidelberg, Germany, Apr. 2007.

[3] R. H. Koenen, J. Lacy, M. Mackay, and S. Mitchell, ``The long march to interoperable digital rights management,’’ Proc. of the IEEE, vol. 92, no. 6, pp. 883--897, Jun. 2004.

[4] C. T. Hsu and J. L. Wu, ``Hidden digital watermarks in images,’’ IEEE Trans. Image Process., vol. 8, no. 1, pp. 58--68, Jan.1999.

[5] B. Macq, J. Dittmann, E. J. Delp, ``Benchmarking of image watermarking algorithms for digital rights management,’’ Proc. of the IEEE, vol. 92, no. 6, pp. 971--984, Jun. 2004.

[6] C. S. Shieh, H.-C. Huang, F. H. Wang, J. S. Pan, ``Genetic watermarking based on transform domain techniques,’’ Patt. Recog., vol. 37, no. 3, pp. 555--565, Mar. 2004.