Research On Audio Digital Watermark Based On Synchronization Technology And Neural Network Theory

With the extensive application and rapid development of the computer networking, Internet has become more and more important way to product, copy and transmit information. However, during we share the change, pirating becomes extremely rampant. World has been aware of the importance of copyright protecting. As one of the techniques of copyright protection, watermarking has been used to protect the copyright of multimedia, such as images, digital audio and video and so on, and becomes active research field in Computer Science. The technique can make sure the ownership of copyright by embedding watermark information into the digital audio. When the copyright is pirated, the author can prove the ownership by watermark in works. Recently, image watermark technique is fairly ripe, but the research on audio watermark is beginning. Digital audio watermark technique is the hot point in watermark research field.This paper focus in digital audio watermark research. Synchronization technique and neural network theory are applied to audio watermark algorithms, respectively.In the anti-cropping audio watermark algorithm based on synchronization technique, the watermark is a binary image. Synchronization code, watermark sequence number and watermark are embedded in discrete cosine transform(DCT) domain of audio signal. The watermark information hiding in audio suffering cropping attack can be recognized by embedding code. The location of watermark in the image can be extracted by embedding watermark sequence number. And the watermark element can be recovered to image by the location information. Furthermore, it is of great benefit to resist the attacks except for cropping that data is embedding in DCT domain. The result of experiments shows that the algorithm have powerful robustness in resisting attacks, especially in anti-cropping.In the audio watermark algorithm based on neural network theory, the watermark is a gray image. Neural network is trained by sample, and the watermark is extracted by the Neural network trained. By the algorithm, a great deal of data can be hided. On the other hand the time of extracting data is short and the anti-attacking robustness is strong. Under attacking, not only data embedded but also sample will suffer the same strength attack. By training neural network, the system can adapt the attack and overcome the influence. The result of experiments prove that the algorithm have strong robustness to resist the noise attacking, re-quantification attacking and so on.