I. INTRODUCTION Watermarking technology has recently received significant attention. This is mainly due to the need to provide copyright protection for digital media such as audio, images and video. Watermarks can be used by holders to identify media buyers for cultural and educational surveillance, determine royalties and copyrights, and determine whether the original data has been altered in some way. The purpose of the latter is somewhat different from copyright control, so its specials will also be different, so it will not be discussed further here.
People study various technologies to avoid being copied illegally. The two best techniques are encoding and watermarking. Encoding protects the data content from the transmission to the receiving transmission. After receiving the decoding, the data will be lost and displayed in a clear manner. The watermark embeds the signal directly in the data to make the coding more perfect. Therefore, the goal of the watermark is to keep the message in the data.
Second, the characteristics of the watermark A watermark product should have a series of desired features, including the imprint is not easy to see, strong anti-signal distortion capability can be decoded by the decoder, anti-intention to remove the ability to imprint, provide the same with the application Data rate, allowing the addition of various imprints, etc. These characteristics are discussed in detail below.
It is not easy to see: the imprint should not be obvious to the observation, and the imprint cannot reduce the quality of the media. For trained observers, the imprint should be perceptible. In addition, significant differences can be perceived by comparing two signals, for example, compressed and uncompressed, watermarked, and primitive.
Integrity: Audio, video, and video signals may be subject to a variety of distortions, but the imprint must be complete during conversion. This conversion includes distortion of the signal distortion, digital-to-analog and analog-to-digital conversion, and lossy compression. For images and video, it is important that the marks can restore geometrical distortions such as conversions, scaling and clipping.
The factual integrity includes two main points: (1) Whether the distorted imprint still exists in the data and (2) whether the imprint can be detected. For example, after the geometric distortion is scaled, the stamps inserted with multiple algorithms are still in the data, but if the distortion is removed first, the corresponding control algorithm can only control the imprint. In this case, either the distortion cannot be determined or the detector cannot detect the imprint.
Anti-deletion capability: The imprint needs to pass through the historical signal processing process to remove the imprint. For the party's signal distortion, the imprint must remain intact. Therefore, the anti-deletion capability of the imprint is important. This can be achieved by private imprinting and asymmetric encoding and decoding.
Bit rate: The bit rate of the imprint is related to the amount of imprints that can be programmed into a letter. This is especially important for the open watermark.
Modifications and multiple watermarks: In some cases, the stamp needs to be changed after insertion. For example, in a digital optical disc, the optical disc may be watermarked to allow only one copy. However, after the copying is completed, it is necessary to change the stamp on the original disc to prohibit further copying. The stamp can be changed in the following way: (1) Remove the first stamp and then add a new stamp. (2) Insert a second imprint so that both are readable, but must be one over another. The first change does not allow the imprinting of the anti-deletion capability because of the emphatic imprint removability. It is a good practice to allow multiple imprints to exist at the same time, which also allows them to insert their own unique imprints from manufacturing to distribution to the final sale.
Restrictive: In commercial applications, the computational cost of encoding and decoding is important, and the importance of encoding cost is to decode. The computational requirements force the watermark to be simple, but this simplicity may significantly reduce the anti-deletion capability. As we all know, the computer's affiliation speed is nearly double every 18 months, so those who seem impossible to calculate today may soon become the Lord. Therefore, it is very necessary to design a stamp. The decoder of this stamp is not limited by the computer age (model number).
Third, the watermark block diagram The process of watermarking an image together can be understood as adding a noise factor, which may be a function of the watermark signal W, or it may be a function of the original image I. The watermark image I can be given by:
I`=I+(I,W)---(1)
The watermark image can withstand a large amount of distortion due to deletion or general use, and this distortion can be expressed as noise. In many cases, noise can be approximated by a linear superposition method. However, distortions in the spatial transformation of an image can be highly nonlinear and are image-dependent, ie n=n(I). Considering the effect of noise, the decoded image is:
I"=I`+n=I+f(I,W)+n(I)--(2)
In decoding, we want to extract the watermark signal W, ie no signal (or noise) is image I at this time. It should be noted that I is much more important than the inserted imprint f(I,W) and distortion n, and the fidelity of the image is preserved. Therefore, the ratio of the signal to the noise input into the decoder (where the signal is now the watermark W) is much less than one. Obviously, using the original image as a part of the verification code procedure can greatly improve the signal to noise ratio. The value can be obtained by simply subtracting the original image I from (2).
Summarizing the characteristics of various proposed watermarking methods, we consider that there are two important characteristics: (1) Whether the watermark is inserted into the visually sensitive area of ​​the image. (2) Whether the inserted signal is independent of the original image I.
Consider the first case where the insertion signal is independent of the original image, ie f(I,W)=W.
In this case, equation (2) simplifies to I"=W+I+N---(3)
Here the signal is slow and the noise is I+N. The chirp signal can be filtered out using the traditional matching method. In this case, if the imprint is placed in the visually sensitive area of ​​the image, then the basic knowledge of the human visual and auditory system is used to imprint the imprint. Must be band-pass filter, as shown in Figure I. However, there is a possible disadvantage in that the formed water money spectrum is independent of the currently known image of the human visual and auditory system. The power in these bands varies greatly from one image to another. Therefore, if the stamps and images formed are simply linearly added, the energy of the imprint must be small to avoid the worst case where the energy of the image is very low in some close-up frequency bands, much lower than the imprint energy, and thus the image Easy to copy. On the contrary, if the image energy is very strong in a close-up band channel, then there is a chance that the country will enter relatively more imprint energy without affecting the image fidelity. (To be continued)
People study various technologies to avoid being copied illegally. The two best techniques are encoding and watermarking. Encoding protects the data content from the transmission to the receiving transmission. After receiving the decoding, the data will be lost and displayed in a clear manner. The watermark embeds the signal directly in the data to make the coding more perfect. Therefore, the goal of the watermark is to keep the message in the data.
Second, the characteristics of the watermark A watermark product should have a series of desired features, including the imprint is not easy to see, strong anti-signal distortion capability can be decoded by the decoder, anti-intention to remove the ability to imprint, provide the same with the application Data rate, allowing the addition of various imprints, etc. These characteristics are discussed in detail below.
It is not easy to see: the imprint should not be obvious to the observation, and the imprint cannot reduce the quality of the media. For trained observers, the imprint should be perceptible. In addition, significant differences can be perceived by comparing two signals, for example, compressed and uncompressed, watermarked, and primitive.
Integrity: Audio, video, and video signals may be subject to a variety of distortions, but the imprint must be complete during conversion. This conversion includes distortion of the signal distortion, digital-to-analog and analog-to-digital conversion, and lossy compression. For images and video, it is important that the marks can restore geometrical distortions such as conversions, scaling and clipping.
The factual integrity includes two main points: (1) Whether the distorted imprint still exists in the data and (2) whether the imprint can be detected. For example, after the geometric distortion is scaled, the stamps inserted with multiple algorithms are still in the data, but if the distortion is removed first, the corresponding control algorithm can only control the imprint. In this case, either the distortion cannot be determined or the detector cannot detect the imprint.
Anti-deletion capability: The imprint needs to pass through the historical signal processing process to remove the imprint. For the party's signal distortion, the imprint must remain intact. Therefore, the anti-deletion capability of the imprint is important. This can be achieved by private imprinting and asymmetric encoding and decoding.
Bit rate: The bit rate of the imprint is related to the amount of imprints that can be programmed into a letter. This is especially important for the open watermark.
Modifications and multiple watermarks: In some cases, the stamp needs to be changed after insertion. For example, in a digital optical disc, the optical disc may be watermarked to allow only one copy. However, after the copying is completed, it is necessary to change the stamp on the original disc to prohibit further copying. The stamp can be changed in the following way: (1) Remove the first stamp and then add a new stamp. (2) Insert a second imprint so that both are readable, but must be one over another. The first change does not allow the imprinting of the anti-deletion capability because of the emphatic imprint removability. It is a good practice to allow multiple imprints to exist at the same time, which also allows them to insert their own unique imprints from manufacturing to distribution to the final sale.
Restrictive: In commercial applications, the computational cost of encoding and decoding is important, and the importance of encoding cost is to decode. The computational requirements force the watermark to be simple, but this simplicity may significantly reduce the anti-deletion capability. As we all know, the computer's affiliation speed is nearly double every 18 months, so those who seem impossible to calculate today may soon become the Lord. Therefore, it is very necessary to design a stamp. The decoder of this stamp is not limited by the computer age (model number).
Third, the watermark block diagram The process of watermarking an image together can be understood as adding a noise factor, which may be a function of the watermark signal W, or it may be a function of the original image I. The watermark image I can be given by:
I`=I+(I,W)---(1)
The watermark image can withstand a large amount of distortion due to deletion or general use, and this distortion can be expressed as noise. In many cases, noise can be approximated by a linear superposition method. However, distortions in the spatial transformation of an image can be highly nonlinear and are image-dependent, ie n=n(I). Considering the effect of noise, the decoded image is:
I"=I`+n=I+f(I,W)+n(I)--(2)
In decoding, we want to extract the watermark signal W, ie no signal (or noise) is image I at this time. It should be noted that I is much more important than the inserted imprint f(I,W) and distortion n, and the fidelity of the image is preserved. Therefore, the ratio of the signal to the noise input into the decoder (where the signal is now the watermark W) is much less than one. Obviously, using the original image as a part of the verification code procedure can greatly improve the signal to noise ratio. The value can be obtained by simply subtracting the original image I from (2).
Summarizing the characteristics of various proposed watermarking methods, we consider that there are two important characteristics: (1) Whether the watermark is inserted into the visually sensitive area of ​​the image. (2) Whether the inserted signal is independent of the original image I.
Consider the first case where the insertion signal is independent of the original image, ie f(I,W)=W.
In this case, equation (2) simplifies to I"=W+I+N---(3)
Here the signal is slow and the noise is I+N. The chirp signal can be filtered out using the traditional matching method. In this case, if the imprint is placed in the visually sensitive area of ​​the image, then the basic knowledge of the human visual and auditory system is used to imprint the imprint. Must be band-pass filter, as shown in Figure I. However, there is a possible disadvantage in that the formed water money spectrum is independent of the currently known image of the human visual and auditory system. The power in these bands varies greatly from one image to another. Therefore, if the stamps and images formed are simply linearly added, the energy of the imprint must be small to avoid the worst case where the energy of the image is very low in some close-up frequency bands, much lower than the imprint energy, and thus the image Easy to copy. On the contrary, if the image energy is very strong in a close-up band channel, then there is a chance that the country will enter relatively more imprint energy without affecting the image fidelity. (To be continued)