A Wavelet-Based Digital Watermarking Algorithm

Digital watermark technology of image is an important branch of information hiding. With the development of internet and multimedia technology, the communication of multimedia information, including digital image, has attained great achievement. As digital image can be easily copied without permit, traditional cryptogram method cannot solve the above problem. This results in great challenge in the area of copyright protection. Digital watermark technology, using signal processing methods, embeds identifier in the digital image. So it is fit to protect the copyright of the digital image, and this is our research backgroundA wavelet-based digital watermarking algorithm is proposed in this paper; the algorithm is based on wavelet transform and tries hard to spread the watermark into various parts of the source image. So the local change of the image cannot destroy the watermark too much. Further more, because DWT (discrete wavelet transform) can transform the image into various frequency bands, we can take advantage of the different ability of resisting attacks of each band, and embed two watermarks into different bands of the wavelet coefficients. So ,the ability of resisting attacks of the algorithm is increased. The algorithm is consist of three parts: watermark modulation,watermark embedding and watermark detecting:一,Watermark ModulationIf we embed the watermark into the source image directly, first the watermark will be easily found and cleared. Second if the image is suffer from cropping , some parts of the watermark will be entirely cropped. If the important part of the watermark (e.g. words) is cropped,the judgment of the copyright will be hard. So we should scramble the watermark before it is embedded into the image. Here, we use the famous Arnold Transform: modN.Here (x, y) is the coordinate of an arbitrary point of the watermark image, and (x',y') is the new coordinate of the point after Arnold Transform. N is the width (height) of the watermark image, notice that the image is square.After transformation, the watermark is scrambled into a binary sequence. Arnold Transform is periodic, after some times of transformation, the image will go back to the origin.二,Watermark EmbeddingIn order to spread the watermark into various parts of the source image, we first divide the source image and the watermark which is scrambled into 8×8 blocks, each block is I_i,j and W_i,j to remember. Then perform two times DWT, and obtains seven frequency bands of wavelet coefficients: LL , LH₂ , HL₂ ,HH₂,LH₁ , HL₁ ,HH₁. Because different band has different robustness to resist attacks, we select the lowest frequency band LL and the middle frequency band LH₂ to embed two watermarks which have different sign. Another advantage of doing this is to reduce the influence to the average luminance of the source image because of the watermark. The order of embedding the watermark is like a snake, during embedding, we choose a balance factorαto compromise the transparency and the intensity of the watermark. Finally, perform two times of reverse DWT to each block of the image, then put the blocks together and obtain image which contain two watermarks. The balance factorαand the times of Arnold transform are saved as keys.The image we obtained at last contains two watermarks, they hasopposite signs, and are equably spread into the image. The watermarks in the lowest frequency and middle frequency can resist different attacks. For instance, the watermark in the lowest frequency can resist Jpeg compression, and the watermark in the middle frequency can resist median filtering. So we can detect two watermarks, and select the watermark which has better quality as the final result.三,Watermark DetectingThe process of detecting the watermarks is the reverse process of embedding. The detecting algorithm in this paper needs the source image.First , divide the image (I|-) which will be detected and source image I into 8×8 blocks, each block is (I|-)_i,j and I_i,j to remember. Then perform two times DWT on each block and and obtains seven frequency bands of wavelet coefficients: LL , LH₂ , HL₂ ,HH₂, and obtains seven frequency bands of wavelet coefficients: LL , LH₂ , HL₂ , HH₂, LH₁, HL₁ ,HH₁. For the lowest frequency and the middle frequency LH 2, we detecting the watermark separately according to the snake shape, and using the formulas which are reverse to the embedding process. Finally we obtain two watermark blocks (W|-)_i,j and W_i,j, which come the lowest and middle frequency. Then put the blocks together and obtain two watermarks (W|-) and W .Notice that the watermarks we obtain here are binary sequence, we need the keys saved before to perform some times Arnold transform to get the binary watermark image. Because the two watermark have different robustness to different attacks, we select the watermark which has better quality as the final result.The experiment shows that the algorithm is robust to common image processing, such as Jpeg compression, median filtering, wiener filtering and cropping. So the algorithm has a certain extent practicability.