Message authentication and digital signature can be applied to digital watermarking.
Traditional watermark is used to prove the legality of the contents on paper money or paper, and digital watermark is used to prove the ownership and authenticity of digital products. Digital watermark is digital information embedded in digital products. It can be the author's serial number, company logo, special text, etc.
Digital watermarking is mainly used to prevent illegal copying (indirectly), determine the ownership (author, publisher, distributor and legal end user), determine the authenticity and integrity of the work (whether it is forged or tampered with), confirm the receiver, undeniable transmission, verification of court evidence, forgery identification, identification of file source and version, Web network inspection and monitoring of thieves, etc.
Traditional watermarking is visible to human eyes, while digital watermarking is divided into perceptual and perceptual types.
Perceptible digital watermarking is mainly used to declare the ownership, copyright and source of products on the spot, and plays the role of advertising or restriction. The detectable watermark is usually a light or translucent pattern that is not ugly; For example, when a TV program is playing, the translucent logo of the TV station is inserted in a corner. Another purpose is to distribute works online, such as giving away a low-resolution image with a visible watermark for free. Watermark is often the information of the owner or seller, which provides clues for finding high-resolution original works. If you want to get high-resolution original works, you have to pay. In order to promote their products on the Internet, some companies first distribute reversible visible watermarks, and then use special software to remove the visible watermarks and add invisible watermarks (publishers, distributors, end users and other information). ) when paying for them. It can be seen that watermark has other uses, that is, to save bandwidth, storage space and other reasons, subtitles and subtitles in multiple languages are embedded in VCD, DVD and other movie copies, and the watermark text in each frame is decoded by hardware in real time and displayed on the screen as needed.
The visible watermark in some products reduces the ornamental value of the works more or less, which makes its use relatively limited. The imperceptible watermark has a higher application level and is more difficult to make.
The imperceptible digital watermark, like invisible words in invisible ink technology, is hidden in digital products. The existence of watermark should be based on the principle of not destroying the appreciation value and use value of original data. The digital watermark is embedded in the protected information in some way. When there is a copyright dispute, the digital watermark is extracted through the corresponding algorithm, so as to verify the ownership of copyright. The protected information can be images, sounds, videos or general electronic documents. In order to make it more difficult for attackers to remove watermarks, most watermarking schemes use keys when embedding and extracting watermarks.
Figure 5.7 Watermark Embedding and Extraction
Although digital watermarking technology can't prevent piracy, it can judge whether the object is protected, monitor the dissemination, authenticity identification and illegal copying of protected data, solve copyright disputes and provide evidence for the court.
The design of digital watermarking needs to consider the following aspects:
Robustness: refers to the ability of protected information to resist the loss of hidden information after some changes. Such as channel noise during transmission, filtering operation, resampling, lossy coding compression, D/ A or A/ D conversion, geometric transformation of images (such as translation, expansion, rotation, cropping, etc.). ).
Undetectability (invisibility) means that the hidden carrier has the same characteristics as the original carrier. Such as consistent statistical noise distribution, it is difficult for illegal interceptors to judge whether there is hidden information.
Transparency: refers to the original data without obvious degradation after a series of hiding processing.
Security: The hiding algorithm is required to have strong anti-attack ability (tampering, forgery and watermark removal), so that the hidden information will not be destroyed. If the watermark is not lost due to file format conversion, and the watermark cannot be detected by unauthorized people.
Self-recovery: Because the original data may be destroyed after some operations or transformations, if the hidden signal is still only recovered from the remaining data fragments, and the original data is not needed in the recovery process, this is self-recovery (self-similarity).
Watermark capacity: Watermark capacity and robustness are contradictory. The increase of watermark capacity will reduce the robustness and also affect the invisibility. In order to resist various transformations, watermarks usually need to be added repeatedly according to a certain arrangement. When the watermark capacity is large, the number of repetitions has to be reduced, and the poor robustness will lead to unreliable detection results.
There are many classifications of digital watermarking technology.
According to different functions, it can be divided into robust watermarking and fragile watermarking. The former is mainly used to mark the copyright information in digital works, and the watermark to be embedded can resist common editing and lossy compression; The latter is mainly used for integrity protection and judging whether the signal has been tampered with.
According to the carrier of watermark, it can be divided into image watermark, video watermark, audio watermark, text watermark and printing watermark.
According to the detection method, it can be divided into transparent watermark and blind watermark. The technology that needs original data in the detection process is called bright watermark, which has strong robustness; The technology that does not need the original data in the detection process is called blind watermarking.
According to the content, it can be divided into content watermark and logo watermark. Content watermarking means that people can still judge the content by feeling after the watermark is destroyed by attack; Marking watermark refers to determining whether there is a watermark mark in the signal through detection and judgment.
According to the use, it can be divided into copyright protection watermark, tampering prompt watermark, bill anti-counterfeiting watermark, hidden logo watermark, printed digital watermark and so on.
The content of digital watermark can be any representative information, such as images, characters, numbers, symbols and so on. In order to hide easily, the smaller the watermark, the better. It is a good choice to use text as watermark information, which can save space and read its meaning directly. Digital watermarking is mainly used in copyright protection, fingerprint identification, title and annotation, tampering tips, use control and other fields.
Copyright protection: that is, the owner of digital media can generate a watermark with a key, embed it in the original data, and then publicly release his watermark version. Digital media includes audio-visual products, digital broadcasting, DVD, MP3 and so on. When a work is pirated or there is a copyright dispute, the copyright owner can obtain the watermark signal from the pirated or watermarked work as a basis, thus protecting the rights and interests of the copyright owner.
Copyright tracing: In order to avoid unauthorized copy making and distribution, the producer can embed the id or serial number of different users in the legal copy of the work as different watermarks, which are called digital fingerprints. Its purpose is to identify the distributed copies of data through the information of authorized users, and to monitor and track the illegal copies during use. Once an unauthorized copy is found, its source can be determined according to the fingerprint recovered from the copy.
Title and comments: The title, comments and other contents of the work are embedded in the work in the form of watermark. This implicit annotation does not need extra bandwidth and is not easy to lose. For example, date, latitude and longitude are hidden in remote sensing images and other information.
Tampering tip: When digital works are used in court, medicine, news and business, it is often necessary to determine whether their contents have been modified, forged or specially treated. In order to achieve this goal, we can usually divide the original image into several independent blocks, and then add different watermarks to each block. At the same time, the integrity of the work can be determined by detecting the watermark signal in each data block. Different from other watermarks, this kind of watermark must be fragile, and the original data is not needed when detecting the watermark signal.
Usage control: A typical example of this application is DVD copy protection system, which adds watermark information to DVD data, so that DVD players can judge its legality and reproducibility by detecting the watermark information in DVD data. So as to protect the commercial interests of manufacturers.
Typical digital watermarking algorithms include the following algorithms.
Spatial algorithm: This algorithm first inputs a key into an M-sequence generator to generate a watermark signal, and then arranges it into a two-dimensional watermark signal, which is embedded into the least significant bit (LSB) of the original image one by one. This ensures that the embedded watermark is invisible. However, due to the use of unimportant pixels in the image, the robustness of the algorithm is poor, and the watermark information is easily destroyed by filtering, image quantization and geometric deformation, so it is not strong enough.
Text watermarking algorithm: watermark is embedded by slightly changing character spacing and line spacing, or adding or deleting character features such as shading lines. Or add a watermark at the symbol level or semantic level, for example, you can replace the big in the text with the big.
Watermarking algorithm based on changing the statistical characteristics of image data: Patchwork algorithm first randomly selects n object pixels, and then reduces the brightness value of one point in the pixel pair by increasing the brightness value of the other point. The average brightness of the whole image remains unchanged. Patchwork method can resist JPEG compression, FIR filtering and image cropping by properly adjusting parameters, but the amount of information embedded in this method is limited.
Frequency domain algorithm: Using one signal can mask the frequency masking phenomenon of another weak signal. A large number of bits can be embedded in the frequency domain space of an image without obvious degradation. When watermarking is added by changing the intermediate frequency or low frequency components (except DC components), the robustness can be greatly improved. Frequency domain watermarking technology can use common transform methods such as discrete cosine transform, wavelet transform and Fourier transform. Its advantages are good hiding effect, and the human eye can't find the difference from the original picture; With key control, only those who know the camouflage key can decrypt it; Can effectively resist cutting and JEPG and other lossy compression coding; The watermark image may be a grayscale image. However, the operation of hiding and extracting information by this algorithm is complicated, and the amount of information hidden cannot be very large.
Compressed domain algorithm: watermark detection and extraction are directly carried out in compressed domain data. A watermark signal is added to a data stream representing a video frame. MPEG-2 data stream can be divided into three parts: header information, motion vector (used for motion compensation) and DCT coded signal block. This algorithm only changes the signal blocks encoded by DCT. Firstly, each input Huffman code in a DCT coded data block is decoded and dequantized to obtain DCT coefficients of the current data block; Secondly, the transform coefficients of the corresponding watermark signal blocks are added to obtain the DCT coefficients of watermark superposition, and then the DCT coefficients are quantized and huffman encoding again. Finally, the bit number n 1 of the new Huffman code word is compared with the original code word n0 without watermark coefficient, and the watermark code word is transmitted only when n 1 is not greater than n0, otherwise, the original code word is transmitted, which ensures that the code rate of the video data stream does not increase. In this method, the introduction of watermark signal is an error signal that causes degradation, and the coding scheme based on motion compensation will spread and accumulate errors. In order to solve this problem, the algorithm adopts a drift compensation scheme to offset the visual distortion caused by the introduction of watermark signal.
NEC algorithm: this algorithm was put forward by COX and others in NEC laboratory, which plays an important role in digital watermarking algorithm. COX believes that the watermark signal should be embedded in the most important part of the source data for human perception. In the spectrum space, this important part is the low frequency component. In this way, in the process of destroying the watermark, the attacker will inevitably cause a serious decline in image quality. Watermark signal should be composed of independent random real number sequences with Gaussian distribution. This greatly enhances the ability of watermark to resist multi-copy joint attack. The implementation method is as follows: firstly, a pseudo-random sequence is generated with the key as the seed, which has a Gaussian n (0, 1) distribution, and the key generally consists of the author's identification code and the hash value of the image. The whole image is transformed by DCT, and the 1000 largest DCT coefficients of the image are modulated (superimposed) by pseudo-random Gaussian sequence. The algorithm has strong robustness, security and transparency.
Physiological model algorithm: The human physiological model includes HVS (Human Visual System) and Human Auditory System (HAS). The basic idea of using visual model is to use JND (just perceptible difference) description derived from visual model to determine the maximum intensity of digital watermark signal that can be tolerated by each part of the image, so as to avoid damaging visual quality. That is to say, the visual model is used to determine the modulation mask related to the image, and then it is used to embed the watermark. This method has good transparency and robustness.
Digital watermarking plays an irreplaceable role in copyright identification, hiding identification and tampering tips, data anti-counterfeiting and so on. It will bring huge commercial profits in commerce, finance, military and personal consumption. Great progress has been made in this field since 1995. With the maturity of digital watermarking technology, digital watermarking technology will play an increasingly important role in e-commerce, video on demand, distance education and distance training.