Intelligent Digital Watermarking Algorithm Based On Chaotic Neural Network

In the twenty-first century, Internet becomes unprecedented grand development. We can see the footprint of it at every corner in our daily life. It has advantage and disadvantage. The Internet was commendable but also digital media can easily be stolen, modified, copied and disseminated, with the attendant problem caused by information security, copyright issues and copyright disputes has become an increasingly serious social problem. Only depending on password protection can not completely effectively solve the problem, then digital watermarking technology comes up. Digital watermarking technology is an important branch of information hide technology, It is mainly through the media tag information embedded in the signal (digital watermark), Provide an effective technical means for media information in the field of anti-counterfeiting, tamper-resistant, certification, protection of data security and integrity.Recently, digital watermarking technology has been a certain amount of development. It has a wide range of algorithm and implementation, but different algorithms have their own deficiencies, how to find an algorithm making the image embedded watermark in all aspects to achieve satisfactory results? How to get the target rapidly, accurately and clearly? Of course, we can continue to optimize the algorithm, or use some existing tools to achieve it. Chaotic neural network is a good instrument. Chaotic neural network not only has the characteristic of chaos such as randomness, regularity, ergodicity of the initial sensitivity and the characteristics of chaotic attractor, but also has the features of neural network such as with massively parallel processing, distributed information storage, self-adaptive self-organization and association study, when the chaotic neural network used in digital watermarking technology, it can not only quickly traverse all the pixels, and can quickly be stable output, but also easy to deal with large amounts of data, it is very feasible. Thus, the algorithm based on chaotic neural network of intelligent digital watermarking technology in this article comes up.This paper presents the development of Digital Watermarking and chaotic neural network, and then introduces the basic concepts related to the watermark, Model of human visual system, wavelet transform theory, as well as the work of chaotic neural network theory. After introducing the design requirements of digital watermark, first describe how to determine the location of embedded watermark,and then, how to select the best watermark strength using chaotic neural network and program. Finally, assess the effect of image and reach a conclusion.The subject of this thesis is focused on chaotic neural networks and intelligent, which is the point and the innovation distinguished from other digital watermarking algorithm. The previous algorithm mainly depends on the use of chaotic neural network, which is sensitive to the initial value of the watermark information to generate uniqueness, so an attacker can not predict the watermark to ensure the information security. When use the neural network model such as BP,Hopfield, it mainly make use of associative memory function to make blind check. In this paper, chaotic neural network is no longer starting from the above two points, but depending on its powerful search optimization features to look for the best embodiment of the watermark embedding strength with watermark invisibility and robustness, which is also the innovation different from other algorithm. The so-called intelligent because of the following reasons: With the development of digital watermarking technology, people have realized that a reasonable choice of the watermark embedding strength plays a vital role to the watermark algorithm. In the past, people select a fixed watermark embedding strength according to the earlier analysis and experimental experience, in recent years, one after another there have been some adjustment about algorithm for watermark embedding strength in accordance with image characteristics, but these algorithms need to re-testing different images and to update some of the parameters, required a great deal of calculation and do not have universal adaptability. In this paper, we use the combination of chaotic neural network and the spatial frequency which reflects the image quality, the changing transiently chaotic neural network energy function is set to negative values of images of the spatial frequency region, as long as the network will begin running, it will begin with chaotic state, and gradually running towards the direction of energy function decreases, when the network reached steady state, the output will be the best watermark embedding strength that the watermark image with the negative space reached minimize of the frequency (spatial frequency reached maximum value). We just need to test well the network parameters at the initial stage of experimental, we can get the final watermark image while any gray image input directly into the network adds a watermark. There's no need to calculate to determine the watermark embedding strength after analyzing the characteristics of different images, which shows the intelligence of the algorithm.Similar with other watermarking algorithm,in this article selection the location of watermark embedding and watermark embedding strength determination are two key steps, But the method of selection is different. About selection the location of watermark embedding, calculation values of the host image after the wavelet decomposition of three-LL3, LH3, HL3, HH3 the matrix based on the DWT domain JND. And rewritten LL3, LH3, HL3, HH3 into a one-dimensional sequence, and in accordance with the size of the corresponding JND value reordering of sequences to be xulie_a, xulie_b, xulie_c, xulie_d; At the same time, obtain one-dimensional sequence in accordance with the result of watermark image after the wavelet decomposition level LL1, LH1, HL1, HH1 the matrix of wavelet coefficients descending sort to be the corresponding. Standing in the front of the wavelet coefficients are embedded in the host image, and this can cause big noise. Because the size of one-dimensional sequence after wavelet transform is 1 * 1024, embed watermark in the first 1024 points at each sequence of xulie_a, xulie_b, xulie_c, xulie_d. This 1024 points corresponds JND threshold, and more watermark information can be embedded, so corresponds with watermark image own better robustness. But watermark information should not be embedded in an unlimited amount, this will affect the invisibility of the watermark. The next key point is to determine the strength of watermark embedding. Chaotic neural network are Chaotic under the status of the energy function and its negative feedback, so network energy function is a critical operation. Setting energy function as negative of a watermark image with spatial frequency region. Network will be stabilized from running continuously through the rough from the chaos-based search to search based on Hopfield thin gradually. At the end, getting maximum output of the watermark image with spatial frequency, that is the best Watermark embedding strength. In order to verify the feasibility of the algorithm, in the experimental part of Chapter 5, we test the robustness in accordance with attacks that digital watermark often encounter. Experiment show that after watermark image with Gaussian noise, salt and pepper noise, shear, rotation, different JPEG quality factor of compression, low-pass, average, median filter, we Still able to extract watermark images from image. These images are more clearly visible on the subjective, Normalized correlation coefficient is above 0.8 objectively. Especially in the rotation and shear attack on the algorithm of the images and data are better than other algorithms. From the view of invisibility, the two sets images we received with a watermark image can not see the existence of watermark, and the values of NC are both above 0.9. On subjective and objective are entirely consistent with the requirements. Therefore, the algorithm is reasonably practicable.