CN1142518C - Method of embedding digital watermark into and separating and recovering digital watermark from media information - Google Patents

Abstract

The present invention relates to a method for inserting digital watermarks into media information and recovering the digital watermarks from the media information, which belongs to the field of mode recognition. Firstly, the two-layer wavelet transformation is carried out on the media information; watermark information and tracking array information are respectively inserted in the coefficient region of H<2>L<2>and L<2>H<2>; inverse wavelet transformation recovers the original media information. After the media information is processed and attacked, the wavelet transformation is firstly carried out on the media information to extract the tracking array; the similarity of the tracking array and the original tracking array is calculated to recover the media information; various attacking and processing methods and different attacking and processing degree are determined according to the similarity of the tracking array; the basis and the reference of the reduced media information are provided; finally, the effective watermark information of the media information is obtained by reducing. After the media information of the present invention is disturbed and processed, the extracted watermark is clear; therefore, the present invention has a good function for distinguishing the pirated and illegally copied media information, etc.

Description

Digital watermarking embedding media information and from media information the method for separating and recovering watermark information

Technical field: the invention belongs to area of pattern recognition, it is specifically related to a kind ofly digital watermarking is hidden in the media information method of this digital watermarking of can extracting out again from media information.

Background technology: digital product is easy to copy so that Internet in vogue today is faced with more stern challenge. Adopting digital watermark technology to carry out copyright protection, is a new focus of digitizing technique in recent years. Have nothing in common with each other in embedding algorithm method and position at present, but common shortcoming is a little less than the anti-attack ability, watermarked media information slightly has interference or conversion process (such as level and smooth, sharpening, scaling, shearing etc.), watermark also changes, rear watermark to be extracted has not been original apperance, do not see original watermark recovery, i.e. watermark is destroyed or lost, this for media information anti-piracy, anti-copy is ineffective.

Summary of the invention: the present invention has developed a kind of digital watermarking embedding media information and the method for separating and recovering watermark information from media information, even the method makes watermarked media information after process general interference and processing (such as level and smooth, sharpening, scaling, shearing etc.), watermark that still can clear effective separating and recovering, and then judge the degree that whether media information is attacked and attacked. The method of watermark information and tracking array information embedding media information is: prepare media information, watermark information and tracking array information, media information is carried out two layers of wavelet transformation, judge whether that all watermark information positions and tracking array information bit all finished embedding, when not being, then select a watermark information position w (m who does not embed, n), read H₂L ₂In coefficient f (m, n), judge w (m, n) whether equal 1, when w (m, n) is not equal to 1, the value of quantization parameter f (m, n) is the value of place section terminal A, as w (m, n) be to equal at 1 o'clock, the value of quantization parameter f (m, n) is the value of place section terminal B, and then selects a tracking array information tc (m who does not embed, n), read corresponding L₂H ₂In coefficient f* (m, n), tracking array is one and has fixed size, and content is complete zero matrix, and require the form of the form of its content and watermark information identical, quantize f* (m, n) value is the value of place section terminal A, then return and judge whether that all watermark information positions and tracking array information bit all finished the embedding program, when all watermark information positions and tracking array information bit have all been finished embedding, media information is carried out inverse wavelet transform, recover media information, telescopiny finishes; The method that media information is quantized is, at first carries out the division of section, and Δ has represented the step-length that quantizes, and A, B 2 class points have represented respectively value and have beenWith

Point, a section namely is take A, B as end points, and long is the interval of Δ, as long as know any one coefficient, just will know that also it drops in a certain section, for H₂L ₂In coefficient, establish it at H₂L ₂In coordinate be (m, n), if the pixel (m of corresponding watermark information, n) value is not equal to 1, then revise the value of this coefficient and be the value of terminal A between its location, if the value of the pixel (m, n) of corresponding binary bitmap is 1, then revises the value of this coefficient and be the value of terminal B between its location. The method that watermark information and tracking array information are separated out from media information is: media information with watermarked information is carried out two layers of wavelet transformation, from L₂H ₂Read coefficient f ' * (m, n) in the section, withCalculate the tracking array content, utilization is distorted estimation function TAF and is calculated the tracking array content separate out and the similarity (1-TAF) of tracking array original contents, media information is carried out inverse wavelet transform reduction media information, select image intensification algorithm commonly used, media information is preserved and backup, media information is processed and is extracted tracking array with selected image intensification algorithm, the tracking array that calculating is extracted and the similarity (1-TAF) of original tracking array content ', is judging that similarity (1-TAF) ' with similarity (1-TAF) is compared improved? if similarity is to be improved, then turning back to media information preserves and stand-by program, if similarity does not improve, then judge whether to also have unworn image intensification algorithm commonly used? if any unworn image intensification algorithm commonly used, then return and select image intensification algorithm routine commonly used, if do not commonly used the image intensification algorithm, then two layers of wavelet transformation are carried out in the new backup of media information, then judge whether to also have still undecomposed watermark information position w* (m, n)? when still undecomposed watermark information position w* (m is arranged, n) time, read H₂L ₂The coefficient f (m, n) of coefficient section utilizes

Separate out still undecomposed watermark information position w* (m, n), and then turn back to and judge whether to also have still undecomposed watermark information position w* (m, n) program, when not still during undecomposed watermark information position w* (m, n), then precipitation process finishes. Among the present invention since the content of tracking array by predefined, so similarity degree of the tracking array tracking array content original with it of extracting out by calculating, just can determine the intensity of variation that it is possible, mode with the backstepping reduction, adopt the method for iteration, both difference is progressively dwindled, when reaching minimum of a value, namely try to achieve one recently like the information of reduction, the number of times of separating out the tracking array iteration has effectively been described the degree that the media information of embed watermark information is tampered, and under the guidance of tracking array, media information is effectively processed, make as much as possible the media information reduction, separate out effect thereby reach better watermark information. Even after media information was interfered and processes, the watermark of extraction was also more clear, this for media information etc. by pirate, illegally copied and had good judgement effect.

Description of drawings: Fig. 1 is the flow chart of watermark information and tracking array information embedding media information, Fig. 2 is the flow chart of separating out watermark information from media information, Fig. 3 is the flow chart that media information is carried out two layers of wavelet transformation, and Fig. 4 is the structural representation that media information is quantized.

The specific embodiment one: this specific embodiment is described is watermark information and tracking array information embedding media information and separates out the method for watermark from media information, the method of watermark information and tracking array information embedding media information is: such as Fig. 1, prepare media information, watermark information and tracking array information, media information is carried out two layers of wavelet transformation, judge whether that all watermark information positions and tracking array information bit all finished embedding, when not being, then select a watermark information position (m who does not embed, n), read H₂L ₂In coefficient f (m, n), judge w (m, n) whether equal 1, when w (m, n) is not equal to 1, the value of quantization parameter f (m, n) is the value of place section terminal A, as w (m, n) equal at 1 o'clock, the value of quantization parameter f (m, n) is the value of place section terminal B, and then selects a tracking array information tc (m who does not embed, n), read corresponding L₂H ₂In coefficient f* (m, n), tracking array is one and has fixed size, and content is complete zero matrix, and require the form of the form of its content and watermark information identical, quantize f* (m, n) value is the value of place section terminal A, then return and judge whether that all watermark information positions and tracking array information bit all finished the embedding program, when all watermark information positions and tracking array information bit have all been finished embedding, media information is carried out inverse wavelet transform, recover media information, embed and finish, the watermark information position that w (m, n) expression does not embed, tc* (m, n) the tracking array information bit separated out of expression, f (m, n) represents the not H of embed watermark information₂L ₂Coefficient, the H behind f ' (m, n) the expression embed watermark information₂L ₂Coefficient, f* (m, n) expression does not embed the L of tracking array information₂H ₂Coefficient. The method that media information is quantized is: such as Fig. 4, in order to reach purpose watermarked in quantization parameter, at first carry out the division of section, Δ has represented the step-length that quantizes, in actual applications, can be according to concrete needs, the integer value of selecting to be fit to is as quantization step; A, B 2 class points have represented respectively value and have been

With

Point, a section namely is take A, B as end points, and long is the interval of Δ, as long as know any one coefficient, just will know that also it drops in a certain section, for H₂L ₂In coefficient, establish it at H₂L ₂In coordinate be (m, n), if the pixel (m of corresponding watermark information, n) value is not equal to 1, then revise the value of this coefficient and be the value of terminal A between its location, if the value of the pixel (m, n) of corresponding binary bitmap is 1, then revises the value of this coefficient and be the value of terminal B between its location. The method that watermark information and tracking array information are separated out from media information is: such as Fig. 2, media information with watermarked information is carried out two layers of wavelet transformation, from L₂H ₂Read coefficient f* (m, n) in the section, withCalculate the tracking array content, utilization is distorted estimation function TAF and is calculated the tracking array content separate out and the similarity (1-TAF) of tracking array original contents, media information is carried out inverse wavelet transform reduction media information, select image intensification algorithm commonly used, media information is preserved and backup, media information is processed and is extracted tracking array with selected image intensification algorithm, the tracking array that calculating is extracted and the similarity (1-TAF) of original tracking array content ', is judging that similarity (1-TAF) ' with similarity (1-TAF) is compared improved? if similarity is to be improved, then turning back to media information preserves and stand-by program, if similarity does not improve, then judge whether to also have unworn image intensification algorithm commonly used? if any unworn image intensification algorithm commonly used, then return and select image intensification algorithm routine commonly used, if do not commonly used the image intensification algorithm, then two layers of wavelet transformation are carried out in the new backup of media information, then judge whether to also have still undecomposed watermark information position w* (m, n)? when still undecomposed watermark information position w* (m is arranged, n) time, read H₂L ₂The coefficient f (m, n) of coefficient section utilizes

Separate out still undecomposed watermark information position w* (m, n), and then turn back to and judge whether to also have still undecomposed watermark information position w* (m, n) program, when not still during undecomposed watermark information position w* (m, n), then precipitation process finishes. The watermark information position that w* (m, n) expression is separated out, the tracking array information bit that tc (m, n) expression does not embed, f* (m, n) expression embeds the L after the tracking array information₂H ₂Coefficient.

The specific embodiment two: this specific embodiment is the method for media information being carried out two layers of wavelet transformation, as shown in Figure 3, the method for two layers of wavelet transformation: media information is carried out obtain low frequency sub-band L behind the LPF h (n) of the LPF h (m) of line direction and column direction₂L ₂, media information is carried out obtain high-frequency sub-band L behind the high-pass filtering g (n) of the LPF h (m) of line direction and column direction₂H ₂, media information is carried out obtain high-frequency sub-band H behind the LPF h (n) of the high-pass filtering g (m) of line direction and column direction₂L ₂, media information is carried out obtain high-frequency sub-band H behind the high-pass filtering g (n) of the high-pass filtering g (m) of line direction and column direction₂H ₂。

The specific embodiment three: what this specific embodiment was described is the algorithm that quantizes: the step-length of Δ for quantizing, value in this specific embodiment is 16, f is the coefficient of intermediate frequency, w is a pixel value of two-value media information, f can be expressed as the form of following formula, and wherein f is the value before quantizing, and fq is the value after quantizing, s is integer, and r is remainder

f＝s×Δ+r    (s∈N，|r|≤Δ)

1)f≥0，w＝1

2)f≥0，w＝0

3)f＜0，w＝1

4)f＜0，w＝0

The specific embodiment four: this specific embodiment is described is that the tracking array content that extracts and the computational methods of original tracking array content similarity are: adopt and distort estimation function TAF and measure the tracking array content tc that extracts^*With the dissimilar degree of tracking array original contents tc, dissimilar degree computing formula is as follows: $\mathrm{TAF}=\mathrm{TAF}(\mathrm{tc},{\mathrm{tc}}^{*})=\frac{1}{M\&times;N}\underset{m=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}[\mathrm{tc}(m,n)\&CirclePlus;{\mathrm{tc}}^{*}(m,\mathrm{nm})]$ M and N represent respectively the number of embedded tracking array row and row, and represents XOR, and M * N represents watermarked total number of bits, value representation tracking array (1-TAF) and the similarity of its original contents.

Claims (4)

1, the method for digital watermarking embedding media information, the method that it is characterized in that watermark information and tracking array information embedding media information is: prepare media information, watermark information and tracking array information, media information is carried out two layers of wavelet transformation, judge whether that all watermark information positions and tracking array information bit all finished embedding, when not being, then select a watermark information position w (m, n) who does not embed, read H₂L ₂In coefficient f (m, n), judge w (m, n) whether equal 1, when w (m, n) is not equal to 1, the value of quantization parameter f (m, n) is the value of place section terminal A, as w (m, n) equal at 1 o'clock, the value of quantization parameter f (m, n) is the value of place section terminal B, and then selects a tracking array information tc (m who does not embed, n), read corresponding L₂H ₂In coefficient f* (m, n), quantize f* (m, n) value is the value of place section terminal A, then return and judge whether that all watermark information positions and tracking array information bit all finished embedding, when all watermark information positions and tracking array information bit have all been finished embedding, media information is carried out inverse wavelet transform, recover media information, telescopiny finishes; The method that media information is quantized is, at first carries out the division of section, and Δ has represented the step-length that quantizes, and A, B 2 class points have represented respectively value and have beenWith

Point, a section namely is take A, B as end points, and long is the interval of Δ, as long as know any one coefficient, just will know that also it drops in a certain section, for H₂L ₂In coefficient, establish it at H₂L ₂In coordinate be (m, n), if the pixel (m of corresponding watermark information, n) value is not equal to 1, then revise the value of this coefficient and be the value of terminal A between its location, if the value of the pixel (m, n) of corresponding binary bitmap is 1, then revises the value of this coefficient and be the value of terminal B between its location.

2, the method for digital watermarking embedding media information according to claim 1, the method that it is characterized in that two layers of wavelet transformation is, media information carried out obtain low frequency sub-band L behind the LPF h (n) of the LPF h (m) of line direction and column direction₂L ₂, media information is carried out obtain high-frequency sub-band L behind the high-pass filtering g (n) of the LPF h (m) of line direction and column direction₂H ₂, media information is carried out obtain high-frequency sub-band H behind the LPF h (n) of the high-pass filtering g (m) of line direction and column direction₂L ₂, media information is carried out obtain high-frequency sub-band H behind the high-pass filtering g (n) of the high-pass filtering g (m) of line direction and column direction₂H ₂。

3, the method for separating and recovering watermark information from media information is characterized in that the method that watermark information and tracking array information are separated out is: media information with watermarked information is carried out two layers of wavelet transformation, from L from media information₂H ₂Read coefficient f ' * (m, n) in the section, with

Calculate the tracking array content, utilization is distorted estimation function TAF and is calculated the tracking array content separate out and the similarity (1-TAF) of tracking array original contents, media information is carried out inverse wavelet transform reduction media information, select image intensification algorithm commonly used, media information is preserved and backup, media information is processed and is extracted tracking array with selected image intensification algorithm, the tracking array that calculating is extracted and the similarity (1-TAF) of original tracking array content ', is judging that similarity (1-TAF) ' with similarity (1-TAF) is compared improved? if similarity is to be improved, then turning back to media information preserves and stand-by program, if similarity does not improve, then judge whether to also have unworn image intensification algorithm commonly used? if any unworn image intensification algorithm commonly used, then return and select image intensification algorithm routine commonly used, if do not commonly used the image intensification algorithm, then two layers of wavelet transformation are carried out in the new backup of media information, then judge whether to also have still undecomposed watermark information position w* (m, n)? when still undecomposed watermark information position w* (m is arranged, n) time, read H₂L ₂The coefficient f ' (m, n) of coefficient section utilizes

Separate out still undecomposed watermark information position w* (m, n), get back to judging whether to also have still undecomposed watermark information position w* (m, n) program, when not still during undecomposed watermark information position w* (m, n), then precipitation process finishes.

4, according to claim 3 from media information the method for separating and recovering watermark information, it is characterized in that the tracking array content that extracts and the computational methods of original tracking array content similarity are: adopt estimation function TAF to measure the tracking array content tc that extracts^*With the dissimilar degree of tracking array original contents tc, dissimilar degree computing formula is as follows: $\mathrm{TAF}=\mathrm{TAF}(\mathrm{tc},{\mathrm{tc}}^{*})=\frac{1}{M\&times;N}\underset{m=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}[\mathrm{tc}(m,n)\&CirclePlus;{\mathrm{tc}}^{*}(m,n)]$ The similarity of value representation tracking array tracking array content original with it (1-TAF).