A watermark embedding method, a watermark extraction method, and a reversible watermark processing system

Abstract

This invention discloses a watermark embedding method, a watermark extraction method, and a reversible watermark processing system, belonging to the field of reversible watermark processing technology. During watermark embedding, a distributed embedding framework is used. As long as a preferred local circular embedding region survives, the watermark sequence can be repositioned and extracted during watermark extraction, fundamentally solving the single-point failure problem. Furthermore, during watermark extraction, different strategies are employed for watermark extraction and image restoration based on whether the final watermark image has suffered external attacks, enabling complete recovery even without external attacks. This invention also improves embedding efficiency and the quality of the final watermark image by embedding compression auxiliary information into the initial watermark image.

Description

Technical Field

[0001] This invention relates to a watermark embedding method, a watermark extraction method, and a reversible watermark processing system, belonging to the field of reversible watermark processing technology. Background Technology

[0002] Current mainstream robust reversible watermarking schemes mostly employ a global embedding strategy, such as embedding Zernike moments or pseudo-Zernike moments based on a global inscribed circle region, to pursue robustness against geometric attacks such as rotation and scaling. However, such methods are extremely vulnerable to clipping attacks or complex geometric attacks involving clipping, because clipping directly destroys the integrity of the watermark carrier, leading to the loss of watermark information and thus a single point of failure problem.

[0003] Meanwhile, existing technologies that increase embedding strength to improve robustness can lead to increased image distortion, a surge in the amount of auxiliary information required for reversible recovery, a large amount of payload, and a reduction in the quality of the watermarked image. Summary of the Invention

[0004] The purpose of this invention is to provide a watermark embedding method, a watermark extraction method, and a reversible watermark processing system to solve the problems of single-point failure and low watermark image quality in the prior art.

[0005] To achieve the above objectives, the present invention employs the following technical solution: In a first aspect, the present invention provides a watermark embedding method, comprising: The original image and watermark sequence are obtained. Multiple original feature points are extracted from the original image. Local circular embedding regions of different sizes are adaptively allocated according to the distance between each original feature point and the center of the original image. Based on each original feature point and each local circular embedding region, multiple preferred local circular embedding regions are obtained by filtering. Polar harmonic Fourier moment coefficients are calculated in each preferred local circular embedding region. The watermark sequence is embedded into each polar harmonic Fourier moment coefficient to obtain the amplitude of each embedded polar harmonic Fourier moment coefficient, and the quantization embedding error and rounding error generated during the embedding process are calculated. The amplitudes of each embedded polar harmonic Fourier moment coefficient are inversely normalized and the pixels are reconstructed to obtain the initial watermark image; Based on the location information of each original feature point, the watermark sequence, the quantization embedding error, and the rounding error, the compression auxiliary information is calculated and embedded into the initial watermark image to obtain the final watermark image.

[0006] Furthermore, the adaptive allocation of local circular embedding regions of different sizes based on the distance between each original feature point and the center of the original image includes: The radius of the corresponding local circular embedding region is calculated based on the Euclidean distance between each original feature point and the center of the original image, thereby adaptively assigning local circular embedding regions of different sizes to each original feature point; The radius of the locally circular embedded region is calculated using the following formula: ; in, Represented as the first The radius of the local circular embedding region assigned to each original feature point. This indicates the maximum radius of the preset central region. This represents the minimum radius of the preset edge region. Indicates the first The Euclidean distance from each original feature point to the center of the original image. This represents half the diagonal length of the original image.

[0007] Furthermore, the step of embedding the watermark sequence into each polar harmonic Fourier moment coefficient to obtain the amplitude of each embedded polar harmonic Fourier moment coefficient includes: Local texture complexity is calculated based on the pixel values ​​of the preferred local circular embedding region; The adaptive normalization factor and quantization step size are calculated based on the order of local texture complexity and polar harmonic Fourier moment coefficients. The amplitude of the polar harmonic Fourier moment coefficients is calculated based on the adaptive normalization factor and the polar harmonic Fourier moment coefficients. Obtain the pseudo-random jitter signal, and construct a quantization function based on the quantization step size and the pseudo-random jitter signal; The quantization function is offset to obtain the first sub-quantizer and the second sub-quantizer. Based on the first sub-quantizer, the second sub-quantizer and the quantization step size, each watermark bit in the watermark sequence is modulated to obtain the amplitude of the embedded polar harmonic Fourier moment coefficient. Perform the above steps on each preferred local circular embedded region and its corresponding polar harmonic Fourier moment coefficient to obtain the amplitude of all embedded polar harmonic Fourier moment coefficients.

[0008] Furthermore, the adaptive normalization factor and quantization step size are calculated based on the order of local texture complexity and polar harmonic Fourier moment coefficients using the following formula: ; in, Represents order The corresponding adaptive normalization factor, Indicates the first The local texture complexity of a preferred local circular embedded region. This indicates the first adjustment parameter. This indicates the second adjustment parameter. This represents the preset adaptive normalization factor; ; in, Indicates the first The order of a preferred local circular embedded region The corresponding quantization step size, Indicates the basic step size. This indicates the third adjustment parameter. This indicates the fourth adjustment parameter.

[0009] Furthermore, the calculation of compression auxiliary information based on the location information of each original feature point, the watermark sequence, the quantization embedding error, and the rounding error includes: Calculate the watermark hash value based on the watermark sequence, quantization embedding error, and rounding error; A location mapping table is generated based on the location information of each original feature point, and a global parameter set is obtained. The global parameter set includes multiple parameters used in the process of embedding the watermark sequence into each polar harmonic Fourier moment coefficient to obtain the amplitude of each embedded polar harmonic Fourier moment coefficient. Calculate the local region error for each preferred local circular embedding region. Based on the local region error, use DPCM predictive coding to calculate the residual between the local region errors of two adjacent preferred local circular embedding regions. Store all residuals in the residual set. The watermark hash value, quantization embedding error, rounding error, position mapping table, global parameter set and residual set are used to form an auxiliary information set. The auxiliary information set is then compressed to obtain compressed auxiliary information with a total bit count that meets the preset requirements.

[0010] Secondly, the present invention provides a watermark extraction method, comprising: Obtain the final watermark image embedded with the watermark sequence and compression auxiliary information, and extract watermark feature points from the final watermark image; Locate the surviving local regions in the final watermarked image, and calculate the surviving polar harmonic Fourier moment coefficients within the surviving local regions; wherein, the surviving local regions are local image patches in the final watermarked image that can still be detected after watermark embedding and / or after being subjected to external attacks; If the final watermark image has not been attacked by external forces, compression auxiliary information is extracted from the final watermark image. The compression auxiliary information is used to compensate the pixels and live polar harmonic Fourier moment coefficients of the final watermark image to obtain the fully restored original image and polar harmonic Fourier moment coefficients. The watermark bits are calculated based on the polar harmonic Fourier moment coefficients, and all watermark bits are merged into a fully restored watermark sequence. If the final watermark image has been attacked, the watermark bits are calculated based on the survival polar harmonic Fourier moment coefficients. The watermark bits corresponding to all the survival polar harmonic Fourier moment coefficients are fused to obtain the incompletely recovered watermark sequence. Compression auxiliary information is extracted from the final watermark image. The compression auxiliary information is used to compensate the surviving local areas and to fill or mark the areas lost due to external attacks, so as to obtain the partially recovered original image.

[0011] Furthermore, if the watermark sequence in the final watermark image is embedded using the following method: Obtain the original image and watermark sequence, and calculate the optimal local circular embedding region and polar harmonic Fourier moment coefficients based on the original image; Local texture complexity is calculated based on the pixel values ​​of the preferred local circular embedding region; The adaptive normalization factor and quantization step size are calculated based on the order of local texture complexity and polar harmonic Fourier moment coefficients. The amplitude of the polar harmonic Fourier moment coefficients is calculated based on the adaptive normalization factor and the polar harmonic Fourier moment coefficients. Obtain the pseudo-random jitter signal, and construct a quantization function based on the quantization step size and the pseudo-random jitter signal; The quantization function is offset to obtain the first sub-quantizer and the second sub-quantizer. Based on the first sub-quantizer, the second sub-quantizer and the quantization step size, each watermark bit in the watermark sequence is modulated to obtain the amplitude of the embedded polar harmonic Fourier moment coefficient. The calculation of the watermark bit based on the survival polar harmonic Fourier moment coefficient is performed using the following formula: ; in, Represents the watermark bits. This indicates taking the minimum value, when When the value is 0, Indicates the first subquantizer, when When taking 1, Indicates the second sub-quantizer. Indicates a pseudo-random jitter signal. Indicates the first The order of the preferred local circular embedding regions is And the number of repetitions is The amplitude of the polar harmonic Fourier moment coefficients This indicates modulo.

[0012] Furthermore, the step of using compression auxiliary information to compensate for the pixels and viable polar harmonic Fourier moment coefficients of the final watermark image to obtain the fully recovered original image and polar harmonic Fourier moment coefficients includes: The surviving polar harmonic Fourier moment coefficients are compensated for by the quantization embedding error in the compressed auxiliary information, and the fully recovered polar harmonic Fourier moment coefficients are obtained. The rounding error in the compression auxiliary information is used to compensate for the pixels of the final watermark image, so as to obtain the fully restored original image.

[0013] Thirdly, the present invention provides a reversible watermarking system, including a watermark embedding end and a watermark extraction end; The watermark embedding end is used to perform the watermark embedding method as described in any one of the first aspects, and send the final watermark image to the watermark extraction end; The watermark extraction end is used to perform the watermark extraction method described in any one of the second aspects.

[0014] Compared with the prior art, the beneficial effects achieved by the present invention are: This invention provides a watermark embedding method, a watermark extraction method, and a reversible watermark processing system. Based on the original feature points, multiple local circular embedding regions are adaptively generated and optimized. The same watermark sequence is embedded in the optimized local circular embedding regions to form a distributed embedding framework. That is, a distributed redundancy strategy is used for watermark embedding. As long as one optimized local circular embedding region survives, the watermark sequence can be relocated and extracted, fundamentally solving the single-point failure problem. By also embedding compression auxiliary information into the initial watermark image, the embedding efficiency and the quality of the final watermark image are improved. Attached Figure Description

[0015] Figure 1 This is a flowchart of a watermark embedding method provided in an embodiment of the present invention; Figure 2 This is a flowchart of a watermark extraction method provided in an embodiment of the present invention. Detailed Implementation

[0016] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to illustrate the technical solution of the present invention more clearly, and should not be used to limit the scope of protection of the present invention.

[0017] Example 1

[0018] like Figure 1 As shown, this embodiment provides a watermark embedding method, including: The original image and watermark sequence are obtained. Multiple original feature points are extracted from the original image. Local circular embedding regions of different sizes are adaptively allocated according to the distance between each original feature point and the center of the original image. Based on each original feature point and each local circular embedding region, multiple preferred local circular embedding regions are obtained by filtering. Polar harmonic Fourier moment coefficients are calculated in each preferred local circular embedding region. The watermark sequence is embedded into each polar harmonic Fourier moment coefficient to obtain the amplitude of each embedded polar harmonic Fourier moment coefficient, and the quantization embedding error and rounding error generated during the embedding process are calculated. The amplitudes of each embedded polar harmonic Fourier moment coefficient are inversely normalized and the pixels are reconstructed to obtain the initial watermark image; Based on the location information of each original feature point, the watermark sequence, the quantization embedding error, and the rounding error, the compression auxiliary information is calculated and embedded into the initial watermark image to obtain the final watermark image.

[0019] This invention adaptively generates multiple local circular embedding regions based on the original feature points and optimizes them. The same watermark sequence is embedded in the optimized local circular embedding regions to form a distributed embedding framework. That is, a distributed redundancy strategy is used for watermark embedding. As long as one optimized local circular embedding region survives, the watermark sequence can be relocated and extracted, which fundamentally solves the single-point failure problem. By also embedding the compression auxiliary information into the initial watermark image, the embedding efficiency and the quality of the final watermark image are improved.

[0020] Example 2

[0021] This embodiment provides a watermark embedding method, including the following steps S1 to S4.

[0022] Step S1: Obtain the original image and watermark sequence, extract multiple original feature points from the original image, adaptively allocate local circular embedding regions of different sizes according to the distance between each original feature point and the center of the original image, filter the local circular embedding regions based on each original feature point and each local circular embedding region to obtain multiple preferred local circular embedding regions, and calculate the polar harmonic Fourier moment coefficients in each preferred local circular embedding region.

[0023] Step S1 can be further divided into steps S1.1 to S1.4.

[0024] Step S1.1: Obtain the original image and watermark sequence, and extract multiple original feature points from the original image.

[0025] Specifically, obtain the original image and watermark sequence. , , Indicates the first Using a watermark bit, the scale-invariant feature transform algorithm is used to extract multiple original feature points from the original image, resulting in an original feature point set.

[0026] Step S1.2: Adaptively allocate local circular embedding regions of different sizes based on the distance between each original feature point and the center of the original image.

[0027] Specifically, to achieve anti-cropping capability, local circular embedding regions of different sizes are adaptively allocated based on the Euclidean distance from each original feature point to the center of the original image.

[0028] No. The coordinates of the original feature points are: The coordinates of the center of the original image are , The width of the original image, expressed in pixels. Representing the height of the original image, then the first... The Euclidean distance from each original feature point to the center of the original image is calculated using the following formula: ; in, Indicates the first The Euclidean distance from each original feature point to the center of the original image. Indicates the first The x-coordinates of the original feature points Indicates the first The ordinates of the original feature points The width of the original image, expressed in pixels. This indicates the height of the original image.

[0029] For the first The radius of the local circular embedding region assigned to each original feature point Calculated using the following formula: ; in, Represented as the first The radius of the local circular embedding region assigned to each original feature point. This represents the maximum radius of the central region, which is set to 150 pixels in this embodiment. This represents the minimum radius of the edge region, which is set to 50 pixels in this embodiment. Indicates the first The Euclidean distance from each original feature point to the center of the original image. This represents half the diagonal length of the original image.

[0030] Step S1.3: Based on each original feature point and each local circular embedding region, the local circular embedding regions are filtered to obtain multiple preferred local circular embedding regions.

[0031] Specifically, to avoid overlapping of local circular embedding regions and control the amount of auxiliary information, this embodiment employs an intelligent selection mechanism to filter the original feature points, resulting in... The preferred original feature points correspond to the preferred local circular embedding regions that are non-overlapping.

[0032] An intelligent selection mechanism is used to filter the original feature points. The specific process is as follows: Step S1.3.1: Calculate the response intensity of all original feature points, i.e., the SIFT response value, and pass it through a threshold. Control the response intensity of the original feature points and remove those with response intensity not exceeding a threshold. The original feature points are used as the remaining original feature points, and the candidate original feature points are initially sorted according to the response intensity from high to low to obtain the candidate original feature point sequence; this step achieves preliminary screening. Step S1.3.2: Select candidate original feature points sequentially from the sorted candidate original feature point sequence, and immediately check whether the local circular embedding region corresponding to the candidate original feature point overlaps with other local circular embedding regions. If they overlap, remove the candidate original feature point; this step achieves further filtering. Step S1.3.3: Sort the candidate original feature points selected in step S1.3.2 according to their Euclidean distance to the center of the original image from smallest to largest, and then take the top ones. The candidate original feature points are selected as the final preferred original feature points, thus obtaining... A non-overlapping preferred local circular embedding region.

[0033] By threshold Controlling the response intensity of the original feature points for intelligent selection A preferred local circular embedding area The value of is generally between 10 and 20, and in this embodiment it is specifically 15. High-quality original feature points near the center of the original image are selected first to ensure that the central content area obtains a larger embedding area, while minimizing the number of local circular embedding areas and controlling the amount of auxiliary information.

[0034] Step S1.4: Calculate the polar harmonic Fourier moment coefficients in each preferred local circular embedding region.

[0035] Specifically, within each preferred local circular embedding region, pixels are mapped to a unit disk and polar harmonic Fourier moment coefficients are calculated.

[0036] The polar harmonic Fourier moment coefficients are calculated using the following formula: ; in, Indicates the first The order of the preferred local circular embedding regions is And the number of repetitions is The polar harmonic Fourier moment coefficients, Indicates the index of the preferred local circular embedded region. This indicates the order of the polar harmonic Fourier moment coefficients. Represents pi (π). Indicates the first A preferred local circular embedding area This indicates that the pixel coordinates in the preferred local circular embedded region will be selected. The pixel values ​​obtained after mapping to the unit disk and normalizing. This represents the x-coordinate of a pixel within the preferred local circular embedded region. This represents the pixel ordinate within the preferred local circular embedded region. Indicates the order is radial basis functions, It does not contain factorials and is computationally efficient. Represents the complex exponential angular direction factor. Represents the imaginary unit. This indicates the number of repetitions of the polar harmonic Fourier moment coefficients. Represents the angle in polar coordinates. Represents the radius in polar coordinates.

[0037] All the polar harmonic Fourier moment coefficients form a set of polar harmonic Fourier moment coefficients, the expression of which is: ; in, express The set of polar harmonic Fourier moment coefficients. Indicates the first The order of the preferred local circular embedding regions is And the number of repetitions is The polar harmonic Fourier moment coefficients, This indicates the order of the polar harmonic Fourier moment coefficients. Indicates the maximum order, used to control capacity and robustness. This indicates the number of repetitions of the polar harmonic Fourier moment coefficients. Indicates the index of the preferred local circular embedded region. This indicates the number of preferred local circular embedding regions. Represents a set of integers.

[0038] Step S2: Embed the watermark sequence into each polar harmonic Fourier moment coefficient to obtain the amplitude of the embedded polar harmonic Fourier moment coefficient, and calculate the quantization embedding error and rounding error generated during the embedding process.

[0039] For the The order of the preferred local circular embedding regions is And the number of repetitions is Polar harmonic Fourier moment coefficients The watermark sequence is embedded with polar harmonic Fourier moment coefficients through adaptive dithering dead zone quantization index modulation to obtain the amplitude of the embedded polar harmonic Fourier moment coefficients, which is specifically expanded into steps S2.1 to S2.8.

[0040] Step S2.1: Based on the preferred local circular embedding region Calculate the local texture complexity of the pixel values. That is, the pixel variance of the preferred local circular embedding region.

[0041] Step S2.2: Based on local texture complexity Harmonic Fourier Moment Coefficients order The adaptive normalization factor is calculated using the following formula: ; in, Represents order The corresponding adaptive normalization factor, Indicates the first The local texture complexity of a preferred local circular embedded region. This indicates the first adjustment parameter. This indicates the second adjustment parameter. The initial adaptive normalization factor is set to 100 in this embodiment, and the balance between robustness and invisibility is optimized through experiments.

[0042] Using the above formula for calculating the adaptive normalization factor, for larger orders... and local texture complexity Larger regions, i.e. regions with complex textures, use larger adaptive normalization factors, allowing for stronger embeddings.

[0043] Step S2.3: Based on the adaptive normalization factor Harmonic Fourier Moment Coefficients The normalized polar harmonic Fourier moment coefficient amplitudes are calculated using the following formula: ; in, Indicates the first The order of the preferred local circular embedding regions is And the number of repetitions is The amplitude of the polar harmonic Fourier moment coefficients Represents order The corresponding adaptive normalization factor, Indicates the first The order of the preferred local circular embedding regions is And the number of repetitions is The polar harmonic Fourier moment coefficients, This indicates modulo.

[0044] Adaptive dithering dead-zone quantization index modulation is used to embed the watermark bits in the watermark sequence into the amplitude of the polar harmonic Fourier moment coefficients. The integer part. This step can be expanded into the following steps S2.4 to S2.7.

[0045] Step S2.4: Based on the order and the Local texture complexity of a preferred local circular embedded region Calculate the quantization step size.

[0046] In this embodiment, the quantization step size of the adaptive jitter dead-zone quantization index modulation process is calculated using the following formula: ; in, Indicates the first The order of a preferred local circular embedded region The corresponding quantization step size, Indicates the basic step size. This indicates the third adjustment parameter. This indicates the fourth adjustment parameter. and The adaptive strength is used to control the adaptive jitter dead zone quantization index modulation process.

[0047] Step S2.5: Introduce a key-controlled pseudo-random jitter signal Based on quantization step size and pseudo-random jitter signal Define the quantization function.

[0048] In this embodiment, with The expression for the quantization function of the independent variable is: ; in, Indicates the first The order of a preferred local circular embedded region The corresponding quantization step size, This represents the rounding function. Indicates a pseudo-random jitter signal. , Indicates the first The order of the preferred local circular embedding regions is And the number of repetitions is The amplitude of the polar harmonic Fourier moment coefficients Indicates the first The order of the preferred local circular embedding regions is And the number of repetitions is The polar harmonic Fourier moment coefficients, This indicates modulo.

[0049] Step S2.6: Set the dead zone width at the center of each quantization interval. , , , This represents the dead zone scaling factor; if the amplitude of the pre-quantization polar harmonic Fourier moment coefficient falls into the dead zone. Use a gentle embedding method or do not embed for the time being.

[0050] in, This represents the quantization index, indicating the center number of the current quantization interval, i.e., the [number]. Each quantized grid represents the nearest integer multiple of the step size center point for that amplitude value, and its calculation formula is as follows: ; in, This represents the rounding function. Indicates the first The order of the preferred local circular embedding regions is And the number of repetitions is The amplitude of the polar harmonic Fourier moment coefficients Indicates the first The order of a preferred local circular embedded region The corresponding quantization step size.

[0051] After introducing the dead zone, with The expression for the quantization function of the independent variable is: ; in, Represents the quantization function. Indicates the first The order of a preferred local circular embedded region The corresponding quantization step size, This represents the rounding function. Indicates a pseudo-random jitter signal. , Indicates the first The order of the preferred local circular embedding regions is And the number of repetitions is The amplitude of the polar harmonic Fourier moment coefficients Indicates the first The order of the preferred local circular embedding regions is And the number of repetitions is The polar harmonic Fourier moment coefficients, To express modulo, This represents the dead zone ratio coefficient.

[0052] Step S2.7: Offset the quantization function in step S2.6 to obtain two sub-quantizers. Based on the two sub-quantizers and the quantization step size in step S2.4, modulate each watermark bit of the watermark sequence to obtain the amplitude of the embedded polar harmonic Fourier moment coefficients.

[0053] The two sub-quantizers are respectively and ,in, Indicates to offset to the left The first sub-quantizer obtained, Indicates to offset to the right The resulting second sub-quantizer.

[0054] The modulation process is performed using the following formula: ; in, Indicates the first The order of the preferred local circular embedding regions is And the number of repetitions is The amplitude of the embedded polar harmonic Fourier moment coefficients Indicates to offset to the left The first sub-quantizer obtained, Indicates to offset to the right The resulting second sub-quantizer, Indicates the first One watermark bit.

[0055] Step S2.8: Calculate the quantization embedding error and rounding error generated in the embedding process in step S2.

[0056] Rounding error is an error generated by the quantizer itself, while quantization embedding error is an error unique to the watermark embedding process. It refers to the additional offset caused by embedding watermark bits. The calculation of these two errors is based on existing technology and will not be elaborated here.

[0057] Step S3: Inversely normalize the amplitudes of each embedded polar harmonic Fourier moment coefficient and reconstruct the pixels to obtain the initial watermark image.

[0058] Step S4: Calculate compression auxiliary information based on the location information of each original feature point, the watermark sequence, the quantization embedding error and the rounding error, and embed the compression auxiliary information into the non-local region or low texture position of the initial watermark image to obtain the final watermark image.

[0059] Step S4 can be further divided into steps S4.1 to S4.7.

[0060] Step S4.1: Calculate the watermark hash value based on the watermark sequence, quantization embedding error, and rounding error. .

[0061] Step S4.2: Generate a location mapping table based on the location information of each original feature point. Get the global parameter set Global parameter set include , , , , , , and , Indicates the initial adaptive normalization factor. This indicates the first adjustment parameter. This indicates the second adjustment parameter. Indicates the basic step size. This indicates the third adjustment parameter. This indicates the fourth adjustment parameter. This represents the dead zone ratio coefficient. This indicates the number of preferred local circular embedded regions.

[0062] Step S4.3: Calculate the local region error for each preferred local circular embedding region. Based on the local region error, use DPCM predictive coding to calculate the residual between the local region errors of two adjacent preferred local circular embedding regions, and store all residuals in the residual set. .

[0063] Step S4.4: Based on the Euclidean distance from each preferred original feature point to the center of the original image, classify the importance of each auxiliary information in steps S4.1 to S4.3. The large central region is recorded and stored with complete high precision, while the small edge region is encoded and stored with lossy quantization.

[0064] Step S4.5: Combine all the above auxiliary information to form an auxiliary information set. , , Indicates quantization embedding error, This indicates the rounding error.

[0065] Step S4.6: Use arithmetic coding or LZMA to set the auxiliary information. Compression, obtaining compression auxiliary information The total number of bits for compressed auxiliary information is controlled to be within 5% of the effective payload.

[0066] Step S4.7: Compress auxiliary information based on classical reversible embedding techniques. The final watermark image is obtained by embedding non-local regions or low-texture locations of the initial watermark image, and the peak signal-to-noise ratio of the final watermark image is greater than 45dB.

[0067] Example 3

[0068] like Figure 2 As shown, this embodiment provides a watermark extraction method, including: Obtain the final watermark image embedded with the watermark sequence and compression auxiliary information, and extract watermark feature points from the final watermark image; Locate the surviving local regions in the final watermarked image, and calculate the surviving polar harmonic Fourier moment coefficients within the surviving local regions; wherein, the surviving local regions are local image patches in the final watermarked image that can still be detected after watermark embedding and / or after being subjected to external attacks; If the final watermark image has not been attacked by external forces, compression auxiliary information is extracted from the final watermark image. The compression auxiliary information is used to compensate the pixels and live polar harmonic Fourier moment coefficients of the final watermark image to obtain the fully restored original image and polar harmonic Fourier moment coefficients. The watermark bits are calculated based on the polar harmonic Fourier moment coefficients, and all watermark bits are merged into a fully restored watermark sequence. If the final watermark image has been attacked, the watermark bits are calculated based on the survival polar harmonic Fourier moment coefficients. The watermark bits corresponding to all the survival polar harmonic Fourier moment coefficients are fused to obtain the incompletely recovered watermark sequence. Compression auxiliary information is extracted from the final watermark image. The compression auxiliary information is used to compensate the surviving local areas and to fill or mark the areas lost due to external attacks, so as to obtain the partially recovered original image.

[0069] Example 4

[0070] This embodiment provides a watermark extraction method, including the following steps S1 to S6.

[0071] Step S1: Obtain the final watermark image that embeds the watermark sequence and compression auxiliary information, i.e., the final watermark image obtained by the method provided in Example 2, and extract watermark feature points from the final watermark image.

[0072] In this embodiment, the watermark feature points are extracted using a scale-invariant feature transformation algorithm.

[0073] Step S2: Locate the surviving local region from the final watermark image, and calculate the surviving polar harmonic Fourier moment coefficients within the surviving local region; wherein, the surviving local region is a local image patch in the final watermark image that can still be detected after watermark embedding and / or after being subjected to external attacks, that is, the preferred local circular embedding region in the final watermark image output in Example 2 that still exists after transmission and being subjected to possible attacks.

[0074] In this embodiment, there is at least one surviving local region. The method for calculating the surviving polar harmonic Fourier moment coefficient in the surviving local region is the same as the method for calculating the polar harmonic Fourier moment coefficient in the preferred local circular embedded region in Embodiment 2, and will not be described again here.

[0075] Step S3: If the final watermark image has not been attacked by external forces, extract compression auxiliary information from the final watermark image, use the compression auxiliary information to compensate the pixels and live polar harmonic Fourier moment coefficients of the final watermark image respectively, and obtain the fully restored original image and polar harmonic Fourier moment coefficients. Calculate the watermark bits based on the polar harmonic Fourier moment coefficients, and fuse all watermark bits into a fully restored watermark sequence.

[0076] In this embodiment, the watermark bits are calculated based on the live polar harmonic Fourier moment coefficients using the following formula: ; in, Represents the watermark bits. This indicates taking the minimum value, when When the value is 0, Indicates the first subquantizer, when When taking 1, Indicates the second sub-quantizer. Indicates a pseudo-random jitter signal. Indicates the first The order of the preferred local circular embedding regions is And the number of repetitions is The amplitude of the polar harmonic Fourier moment coefficients This indicates modulo.

[0077] The pixels and viable polar harmonic Fourier moment coefficients of the final watermark image are compensated using compression auxiliary information to obtain the fully recovered original image and polar harmonic Fourier moment coefficients, including: The quantization embedding error in the compressed auxiliary information is used to accurately inversely compensate the surviving polar harmonic Fourier moment coefficients to obtain the fully recovered polar harmonic Fourier moment coefficients, which are completely consistent with those in Example 2. Thus, the watermark sequence is recovered 100% accurately, and is the same as the watermark sequence obtained in Example 2. The rounding error in the compression auxiliary information is used to compensate for the pixels of the final watermark image, so as to obtain the fully restored original image.

[0078] Step S4: If the final watermark image has been attacked by external forces, calculate the watermark bits based on the live polar harmonic Fourier moment coefficients, perform majority voting to fuse the watermark bits corresponding to all live polar harmonic Fourier moment coefficients to obtain the incompletely recovered watermark sequence, extract compression auxiliary information from the final watermark image, use the compression auxiliary information to compensate for the live local areas, and fill or mark the areas lost due to external attacks to obtain the partially recovered original image.

[0079] Example 5

[0080] This embodiment provides a reversible watermarking system, including a watermark embedding end and a watermark extraction end; The watermark embedding end is used to execute the watermark embedding method provided in Example 1 and send the final watermark image to the watermark extraction end; The watermark extraction end is used to execute the watermark extraction method provided in Example 3.

[0081] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0082] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0083] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0084] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0085] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A watermark embedding method, characterized in that, include: The original image and watermark sequence are obtained. Multiple original feature points are extracted from the original image. Local circular embedding regions of different sizes are adaptively allocated according to the distance between each original feature point and the center of the original image. Based on each original feature point and each local circular embedding region, multiple preferred local circular embedding regions are obtained by filtering. Polar harmonic Fourier moment coefficients are calculated in each preferred local circular embedding region. The watermark sequence is embedded into each polar harmonic Fourier moment coefficient to obtain the amplitude of each embedded polar harmonic Fourier moment coefficient, and the quantization embedding error and rounding error generated during the embedding process are calculated. The amplitudes of each embedded polar harmonic Fourier moment coefficient are inversely normalized and the pixels are reconstructed to obtain the initial watermark image; Based on the location information of each original feature point, the watermark sequence, the quantization embedding error, and the rounding error, the compression auxiliary information is calculated and embedded into the initial watermark image to obtain the final watermark image.

2. The watermark embedding method according to claim 1, characterized in that, The adaptive allocation of local circular embedding regions of different sizes based on the distance between each original feature point and the center of the original image includes: The radius of the corresponding local circular embedding region is calculated based on the Euclidean distance between each original feature point and the center of the original image, thereby adaptively assigning local circular embedding regions of different sizes to each original feature point; The radius of the locally circular embedded region is calculated using the following formula: ； in, Represented as the first The radius of the local circular embedding region assigned to each original feature point. This indicates the maximum radius of the preset central region. This represents the minimum radius of the preset edge region. Indicates the first The Euclidean distance from each original feature point to the center of the original image. This represents half the diagonal length of the original image.

3. The watermark embedding method according to claim 1, characterized in that, The step of embedding the watermark sequence into each polar harmonic Fourier moment coefficient to obtain the amplitude of each embedded polar harmonic Fourier moment coefficient includes: Local texture complexity is calculated based on the pixel values ​​of the preferred local circular embedding region; The adaptive normalization factor and quantization step size are calculated based on the order of local texture complexity and polar harmonic Fourier moment coefficients. The amplitude of the polar harmonic Fourier moment coefficients is calculated based on the adaptive normalization factor and the polar harmonic Fourier moment coefficients. Obtain the pseudo-random jitter signal, and construct a quantization function based on the quantization step size and the pseudo-random jitter signal; The quantization function is offset to obtain the first sub-quantizer and the second sub-quantizer. Based on the first sub-quantizer, the second sub-quantizer and the quantization step size, each watermark bit in the watermark sequence is modulated to obtain the amplitude of the embedded polar harmonic Fourier moment coefficient. Perform the above steps on each preferred local circular embedded region and its corresponding polar harmonic Fourier moment coefficient to obtain the amplitude of all embedded polar harmonic Fourier moment coefficients.

4. The watermark embedding method according to claim 3, characterized in that, The adaptive normalization factor and quantization step size are calculated based on the order of local texture complexity and polar harmonic Fourier moment coefficients, using the following formula: ； in, Represents order The corresponding adaptive normalization factor, Indicates the first The local texture complexity of a preferred local circular embedded region. This indicates the first adjustment parameter. This indicates the second adjustment parameter. This represents the preset adaptive normalization factor; ； in, Indicates the first The order of a preferred local circular embedded region The corresponding quantization step size, Indicates the basic step size. This indicates the third adjustment parameter. This indicates the fourth adjustment parameter.

5. The watermark embedding method according to claim 1, characterized in that, The compression auxiliary information calculated based on the location information of each original feature point, the watermark sequence, the quantization embedding error, and the rounding error includes: Calculate the watermark hash value based on the watermark sequence, quantization embedding error, and rounding error; A location mapping table is generated based on the location information of each original feature point, and a global parameter set is obtained. The global parameter set includes multiple parameters used in the process of embedding the watermark sequence into each polar harmonic Fourier moment coefficient to obtain the amplitude of each embedded polar harmonic Fourier moment coefficient. Calculate the local region error for each preferred local circular embedding region. Based on the local region error, use DPCM predictive coding to calculate the residual between the local region errors of two adjacent preferred local circular embedding regions. Store all residuals in the residual set. The watermark hash value, quantization embedding error, rounding error, position mapping table, global parameter set and residual set are used to form an auxiliary information set. The auxiliary information set is then compressed to obtain compressed auxiliary information with a total bit count that meets the preset requirements.

6. A watermark extraction method, characterized in that, include: Obtain the final watermark image embedded with the watermark sequence and compression auxiliary information, and extract watermark feature points from the final watermark image; Locate the surviving local regions in the final watermarked image, and calculate the surviving polar harmonic Fourier moment coefficients within the surviving local regions; wherein, the surviving local regions are local image patches in the final watermarked image that can still be detected after watermark embedding and / or after being subjected to external attacks; If the final watermark image has not been attacked by external forces, compression auxiliary information is extracted from the final watermark image. The compression auxiliary information is used to compensate the pixels and live polar harmonic Fourier moment coefficients of the final watermark image to obtain the fully restored original image and polar harmonic Fourier moment coefficients. The watermark bits are calculated based on the polar harmonic Fourier moment coefficients, and all watermark bits are merged into a fully restored watermark sequence. If the final watermark image has been attacked, the watermark bits are calculated based on the survival polar harmonic Fourier moment coefficients. The watermark bits corresponding to all the survival polar harmonic Fourier moment coefficients are fused to obtain the incompletely recovered watermark sequence. Compression auxiliary information is extracted from the final watermark image. The compression auxiliary information is used to compensate the surviving local areas and to fill or mark the areas lost due to external attacks, so as to obtain the partially recovered original image.

7. The watermark extraction method according to claim 6, characterized in that, If the watermark sequence in the final watermark image is embedded using the following method: Obtain the original image and watermark sequence, and calculate the optimal local circular embedding region and polar harmonic Fourier moment coefficients based on the original image; Local texture complexity is calculated based on the pixel values ​​of the preferred local circular embedding region; The adaptive normalization factor and quantization step size are calculated based on the order of local texture complexity and polar harmonic Fourier moment coefficients. The amplitude of the polar harmonic Fourier moment coefficients is calculated based on the adaptive normalization factor and the polar harmonic Fourier moment coefficients. Obtain the pseudo-random jitter signal, and construct a quantization function based on the quantization step size and the pseudo-random jitter signal; The quantization function is offset to obtain the first sub-quantizer and the second sub-quantizer. Based on the first sub-quantizer, the second sub-quantizer and the quantization step size, each watermark bit in the watermark sequence is modulated to obtain the amplitude of the embedded polar harmonic Fourier moment coefficient. The calculation of the watermark bit based on the survival polar harmonic Fourier moment coefficient is performed using the following formula: ； in, Represents the watermark bits. This indicates taking the minimum value, when When the value is 0, Indicates the first subquantizer, when When taking 1, Indicates the second sub-quantizer. Indicates a pseudo-random jitter signal. Indicates the first The order of the preferred local circular embedding regions is And the number of repetitions is The amplitude of the polar harmonic Fourier moment coefficients This indicates modulo.

8. The watermark extraction method according to claim 6, characterized in that, The step of using compression auxiliary information to compensate for the pixels and viable polar harmonic Fourier moment coefficients of the final watermark image to obtain the fully recovered original image and polar harmonic Fourier moment coefficients includes: The surviving polar harmonic Fourier moment coefficients are compensated for by the quantization embedding error in the compressed auxiliary information, and the fully recovered polar harmonic Fourier moment coefficients are obtained. The rounding error in the compression auxiliary information is used to compensate for the pixels of the final watermark image, so as to obtain the fully restored original image.

9. A reversible watermarking system, characterized in that, Includes the watermark embedding end and the watermark extraction end; The watermark embedding end is used to perform the watermark embedding method according to any one of claims 1 to 5, and send the final watermark image to the watermark extraction end; The watermark extraction end is used to perform the watermark extraction method according to any one of claims 6 to 8.

Images