DWT-SVD Combination Method for Copyrights Protection

Many irresponsible people who download other people's artwork then modify and redistribute the work of someone without permission by not including the owner's name. Based on that problem an attempt was made to provide an invisible watermark so as not to be easy to do the watermark removal process on the artwork, using a 2-level DWT and SVD combination method that can embed watermarks invisibly and increase watermarking resistance to the image from various image manipulation processes such as noise, JPEG compression, and so on. From combination method of DWT-SVD got the highest PSNR value with watermark image size 32x32 and 64x64 which is 59.2656 dB and 54.7414 dB, respectively.


INTRODUCTION
The development of technology, especially in the field of the internet, is increasing rapidly [1] [2], facilitate some people in accessing something practically on the internet [3]. People often use the internet as a social media to communicate with each other without having to meet face-to-face, disseminate information or spread a work of art such as digital painting, original music, animated videos, and so forth. The easier accessing internet, many irresponsible people who download the artwork of others then modify [4], and redistribute the work without permission and does not include the owner's name that caused the loss to the owner in case of misuse of the dissemination done illegally [5]. So, we make an effort to provide watermarks on the artwork.
The visible watermark [6] is good to know directly who the owner of the artworks but in that way, it will degrade the quality of work because it will be covered by the watermark and make irresponsible people to easily remove the watermark in various ways, whereas if watermarking is done invisibly [4] the work will not be covered and confuse irresponsible people who want to remove the watermark from the artwork.
According to [7], embedding the invisible watermark can be grouped into 2 ways, which are the spatial domain and the transform domain. The spatial domain can embed the watermark directly into the host image pixel and can be implemented easily and quickly, while in the transform domain, the hots image is converted to the frequency form before embedding watermark [8]. The domain transform method, robust against various kinds of image manipulation processes such as compression process [2], therefore, in this paper use domain transform method.
One of the transform domain methods used in this paper is the combination of the SVD (Singular Value Decomposition) method and the 2-level DWT (Discrete Wavelet Transform) method. After method has been applied, it has been tested with various attacks such as salt and pepper, Gaussian filter and JPEG compression and analyzed using PSNR (Peak to Signal Noise Ratio) calculation to measure imperceptibility value and NC calculation (Normal Correlation) to measure robustness value [9].

Digital Watermarking
The watermarking technique works by embedding a message that can be text, image, or audio that shows ownership of digital media without degrading image quality [10]. The goal is to embed an invisible watermark image and measure the imperceptibility criteria. Good or bad criteria on watermarking can be known by using the calculation of PSNR (Peak Signal to Ratio) and see the value of equation (1) and equation (2) [11].
The formula for embedding the watermark image process into the host image: and the extraction process formula to get the watermark image from the watermarked image.
Where, = host image = watermarked image = coefficient multiplier to determine the strength of the watermark image to be embedded = watermark image = extracted watermark image.

2D-Discrete Wavelet Transform
The DWT method can be used for image transformation and image compression. In addition to image processing, the DWT method can also be applied to the steganography and watermarking fields [12]. The process of the wavelet transforms conceptually simple. The original image is decomposed into 4 new sub-images LL, LH, HL, and HH [13]. Figure 1. DWT Sub-images [10] In Figure 1, the sub-image in HL, LH, and HH positions will look like a rough version of the host image because it contains high-frequency components of the host image, while for the LL sub-image looks like the host image and looks smoother as it contains low-frequency components of the host image [14]. Since it is similar to the host image, the LL sub-image can embed the watermark image to the host image [6]. While the pixel values of LH, HL, and HH sub-images tend to be low-value and sometimes zero, so they are easy to compress.

Singular Value Decomposition
The result of the SVD method is the singular value stored in a diagonal matrix, S, in the appropriate order from high to low [13]. Where the value of the singular value of an image has good stability when given a little disturbance will not be too influential [15]. If I is a real m x n size matrix, it is decomposed using the SVD method and will produce the matrices U, S, and V, where the matrices U and V are orthogonal singular vectors and S is the diagonal vector that stores the singular value of the matrix I [16].

Noise Salt and Pepper (Impulse Noise)
Noise on the image can occur either because of pixel dysfunction in the digital camera sensor [16]. In grayscale, the noise will appear as salt and pepper that is white if the grayscale value is 255 and black if the grayscale value is 0.

Filter Gaussian
Gaussian filters are helpful in terms of reducing noise, high-frequency effects, or softening edges in an image [17]. In this paper, Gaussian filter is used to test the robustness of the DWT-SVD combination method.

JPEG Compression
JPEG (Joint Photographics Extent Group) is a standard digital image compression scheme. There are two methods in JPEG compression, which are lossy and lossless methods. The lossy compression method results in a higher compression ratio than the lossless method [18].

Watermarking Evaluation
In this paper will use PSNR calculations as an evaluation tool to measure the quality between the original image and the watermark image [19] and the NC calculation to measure the similarity of the watermarked extraction image. The NC value ranges from 0 -1, where if the watermarked extraction value is close to 1 or equal to 1 then it is similar to the watermark image as shown in equation (3) until equation (5).

Proposed Method using DWT-SVD
In this paper DWT transformation performed 2 level on the host grayscale image, then decomposed SVD and pasted on the diagonal vector. The watermarking scheme as illustrated in Figure 2 and Figure 3. As for the extraction process of watermark image using non-blind method that requires the host image to perform the extraction process. Here's the watermark extraction scheme on the grayscale image as shown in Figure 3. 1) The grayscale watermark image (Iw) is decomposed with 2-level DWT method on the LL sub-image to Iw, LH2, HL2, and HH2 2) Then Iw image decomposed with SVD method into matrix U, Sw, and V 3) After that reconstruct the image from the Uw, Sw, and Vw matrices 4) Then extract the watermark image from Se 5) Extracted watermark image is obtained After obtaining the watermarked image and the watermark image of the extraction result, the image will be tested with the various attack from giving of noise salt and pepper, Gaussian filter, and JPEG Compression with a lossy method. Then we will calculate the PSNR value from the watermarked image and calculate the NC value of the watermarked image of the extraction result.  Based on Table 1, we can see that the PSNR value of the DWT method is clearly less good than the SVD method and the DWT-SVD combination. In Figure 1 we can see the PSNR value is 54.7414 dB, in Table 7 the PSNR value is 54.6680 dB, and Table 3 the PSNR value is 39.9218 dB.   As describe in Figure 6 and Figure 7, the lowest PSNR value is the DWT method. In Figure 6 and Figure 7, the lowest average NC value with the attack Gaussian filter is the DWT method and the highest average NC value with the attack gaussian filter is the DWT-SVD combination method. In this paper, experiment has been tested and compared the PSNR values between the 3-level DWT-SVD combination method and the 2-level DWT-SVD combination method with the 32x32 watermark image size. The PSNR value comparison results in Figure 8. Figure 10. The PSNR value comparison results between the 2-level DWT-SVD combination method and the 3-level DWT-SVD combination method with the 32x32 watermark image size According to Figure 10, the PSNR value between the 3-level DWT-SVD combination method is higher than the 2-level DWT-SVD combination method, where the highest value is 59,4601 dB.

CONCLUSION
From the results of the research, it can be concluded that the average DWT-SVD combination method produces high PSNR values rather than SVD or DWT methods, making it hard to see the watermark image visually. The DWT-SVD combination method also proves to be resistant to JPEG compression, noise, and Gaussian filters. L e n a . p n g B a b o o n . p n g F r u i t s . p n g P e p p e r s . p n g C a t . p n g DWT-SVD 2-level DWT-SVD 3-level