Digital Watermarking Technique Based On Wavelet Packet Transform

Digital watermark, as an effective method for the copyright protection of multimedia, has drawn extensive attention in recent years. There are so many digital watermark algorithms at present. Algorithms especially based on discrete wavelet transform (DWT) are the hotspot. As the further development of wavelet, wavelet packet transform (WPT) compensates for the shortcomings that wavelet can't decompose the high frequency band, and enormously develops the space of watermark information embedded. So the method for watermark algorithms based on WPT has been paid more attention. Although more and more watermark embedding algorithms based on WPT have appeared in the field of image watermarks at present, most of these methods use pseudo-random sequences or binary images as watermarks embedded, involved less gray images, and most of the researches in this field are based on single watermark.Two image watermarking algorithms based on WPT are proposed in this paper, for the insufficiency of watermark algorithms in wavelet packet domain. First, an algorithm based on larger gray-level image is proposed for the deficiency in abundant information of watermarks embedded in wavelet packet domain, namely a gray-level image watermark algorithm based on WPT is implemented. Then, a Bi-watermarking algorithm of gray-level image watermark and binary image watermark based on WPT is proposed, for the relatively poor application and security of single watermark and unblind watermark in wavelet packet domain. The main contributions in this paper are as follows.In the algorithm of single gray-level watermark in wavelet packet domain, the host image is decomposed by the three-layer best wavelet packet transform, and the entropy-optimization algorithm is used to select the best wavelet basis. The host image feature values are obtained by calculating leaf nodes'energy values of the optimal tree. These few and accurate feature points can represent images, which greatly increase processing speed of the system. Before embedded, watermark image is divided into four blocks, and the decomposition way is saved as a key, which increases the difficulty of extracting watermark to attackers. The wavelet packet coefficients of embedding regions are further decomposed using discrete cosine transform (DCT), which removes the correlation among the adjacent coefficients of sub-band images embedded, and reduces the visual distortion. Compared with traditional watermark algorithms based on WPT, it is a great breakthrough in the kinds and amount of watermark information. Under most attack tests of robust performance, the comparison experiments by Matlab show the greater normalized correlation coefficients of the watermarks extracted, which results in better performance, compared with traditional watermark algorithm based on WPT.In the Bi-watermark algorithm of wavelet packet domain, the larger gray-level image watermark and binary watermark are embedded simultaneously. A BP neural network is used to build the extracting model of the binary watermark. The 3*3 square window around embedded points is selected as the input signal of the BP network. With the increment of diagonal coefficients, the information around embedded points is more comprehensive. Therefore, the 3*3 square window is superior to the traditional cross window. The entire extraction process does not require the original image, that is to say, the blind extraction is realized in this algorithm. The dual watermarks embedded greatly enrich the amount of watermark information, which leads to better stability of watermark. Under the robust attacks, both the gray watermark and blind extracting binary watermark have more excellent performance indicators, and the watermarks extracted can be seen clearly by eyes. The experiment results prove the feasibility of this algorithm.