A robust image watermarking method based on moire pattern to resist screen capture

Abstract

The application discloses a robust image watermarking method based on moire mode to resist screen shooting, which encodes watermark information in the carefully designed image moire fringe. By using the embedding mode of superposition of high-frequency subbands in the DTCWT domain and different angle grating templates, the carrier presents "moire mode" in the spatial domain. In addition, the effective blind extraction of the watermark is realized by combining the angle-based decoding method. The comprehensive experiments under different shooting distances, angles, illumination conditions and various shooting and display devices show that the application exhibits good robustness in complex scenes and can effectively resist the interference and distortion in the screen shooting process.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of robust watermarking technology in image information hiding, which is used for image copyright protection and tracing in the screen shooting scenario. BACKGROUND

[0002] Traditional robust image watermarking algorithms have been extensively studied for electronic channel distortion. To cope with image processing distortion in the electronic channel, most methods embed watermark information by finding invariants in the spatial or frequency domain. With the widespread popularity of electronic devices, cross-channel transmission of images has become a research hotspot. The main cross-channel transmission methods currently include the print-scan channel, the print-shooting channel, and the screen-shooting channel. An effective watermarking scheme that can resist cross-channel transmission is based on spatial template design, such as the literature "Screen-Shooting Resilient Watermarking" published in IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY in 2019, Volume 14, Issue 11, pp. 1403-1418. To resist noise during the screen-shooting process, researchers first find the most stable feature points in the carrier and embed the watermark in the surrounding area of these feature points. In addition, some methods embed watermark information based on stable coefficients in the frequency domain, such as the literature "Efficient General Print-Scanning Resilient Data Hiding Based on Uniform Log-Polar Mapping" published in IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY in 2010, Volume 5, Issue 5. In recent years, watermarking methods based on deep learning have received extensive attention. With its powerful learning ability, it has achieved good results. Researchers decompose the distortion in the real cross-channel transmission process into a series of consecutive image processing steps, and then construct a noise layer to improve its robustness to cross-channel attacks, such as the literature "Stegastamp: Invisible hyperlinks in physical photographs" published in Proceedings of the IEEE / CVF conference on computer vision and pattern recognition in 2020, pp. 2117-2126. However, deep learning relies on data sets and is often limited by the size of the image resolution, which contradicts the current high-pixel imaging devices. Due to the inherent characteristics of the distortion in the cross-channel transmission process, such as the moire distortion in the screen-shooting process, which is caused by the interaction between the screen layer and the camera sensor layer, resulting in a moire pattern similar to "water waves". In recent years, some scholars have used "moire patterns" to design a traceability scheme.For example, the paper "mID: Tracing Screen Photos via Moire Patterns" published in 2021 at the 30th USENIX Security Symposium, number 978-1-939133-24-3. But the "Moiré pattern" generation method of this method depends on the sensor of the capture device, such as the CMOS sensor of the camera, and the camera module integrated in the smartphone. The resolution, photosensitive performance and optical characteristics of these sensors are significantly different, and these differences directly affect the generation effect of the Moiré pattern. Therefore, in the actual application environment, due to the diversity of shooting devices (such as professional cameras, smartphone cameras, drone cameras, etc.) and the variability of applicable scenarios (shooting distance, angle, lighting conditions), the accuracy of the "Moiré pattern" is directly affected. The present application proposes a watermarking scheme, which makes the watermark itself present in the form of "Moiré pattern", thereby realizing a Moiré pattern-based anti-screen shooting robust image watermarking method. This method can significantly resist interference and distortion in the screen shooting process, effectively improving the accuracy of watermark information decoding. SUMMARY

[0003] In view of the shortcomings of the prior art, the purpose of the present application is to design a screen shooting robust image watermarking method in which the watermark signal itself is a "Moiré pattern". In the cross-channel image watermark screen shooting distortion, the watermark signal is distinguished, and the shooting device and the environment have wide adaptability, which can significantly resist interference and distortion in the screen shooting process, effectively improving the accuracy of watermark information decoding.

[0004] The technical solution for achieving the purpose of the present application is as follows:

[0005] A robust image watermarking method based on Moiré pattern to resist screen shooting, including embedding process and extraction process;

[0006] The embedding process includes the following steps:

[0007] Step 1: generate an encoded watermark sequence from the original message sequence through watermark preprocessing;

[0008] Step 2: obtain different angle raster templates based on angle encoding of the encoded watermark sequence of step 1;

[0009] Step 3: for the original image, read its RGB channel, and perform a 2-level dual-tree complex wavelet transform (DTCWT) on each channel to obtain a series of high-frequency subbands;

[0010] Step 4: based on the strong correlation of part of the high-frequency subbands, superimpose the raster template into the high-frequency subbands obtained in step 3; ​

[0011] Step 5: Perform inverse DTCWT transform on the high frequency subbands after superimposing the grating template in step 4 to obtain the RGB channel containing the watermark, then merge the three channels to obtain the final image containing the watermark.

[0012] The extraction process includes the following steps:

[0013] Step A: Perform perspective transformation on the captured image to obtain the corrected image.

[0014] Step B: Decompose the corrected image into RGB three channels, and perform DTCWT transform on each channel to obtain a series of high frequency subbands containing angle grating templates.

[0015] Step C: Preprocess the high frequency subbands containing angle grating templates, i.e. perform local histogram equalization to enhance the grating template signal and thus enhance the angle feature.

[0016] Step D: Perform angle-based decoding on the preprocessed high frequency subbands to obtain the extracted watermark information.

[0017] Further, in step 1, the message sequence is processed through watermark preprocessing to generate the encoded watermark sequence, the specific process being:

[0018] For a message sequence m of length l m , perform BCH encoding to obtain a sequence w of length l w . Then w is n-bit Gray code encoded to finally obtain the Gray code sequence where s i is composed of n bits, and k = l w / n.

[0019] Further, in step 2, the encoded watermark sequence is angle-encoded to obtain grating templates of different angles, the specific process being:

[0020] (1) Construct a mapping function p from Gray code to angle:

[0021]

[0022] This formula maps the n-bit Gray code sequence where each g i is composed of n bits, to the corresponding angle set Specifically, this means dividing 180° into 2 n angle intervals, and each Gray code g i is mapped

[0023] ​​Center angle θ of the belonging angle interval i :

[0024]

[0025] where Δθ represents the size of each angle interval, defined as:

[0026] (2) Construct the mapping function q from angle to grating template:

[0027]

[0028] The above formula maps to grating template where l i can be expressed as:

[0029] l i (x,y)=p(φ i (x,y)),

[0030] p(u)=0.5+0.5cos(2πf·u),

[0031] φ i (x,y)=xcosθ i +ysinθ i .

[0032] Here f represents the frequency of the grating, and then the composite mapping can be obtained:

[0033]

[0034] Based on the composite mapping, the Gray code sequence can be mapped to the grating template set

[0035] Further, in step 3, the RGB channels of the image are read, and a level Dual-Tree Complex Wavlet Transform (DT CWT) transform is performed on each channel, and then a series of high-frequency subbands are obtained, and the specific process is as follows:

[0036] The host image I is decomposed into RGB three channels to obtain I R , I G , and I B , and a level DT CWT decomposition is performed, and 6 high-frequency signals (H) and one low-frequency signal can be obtained for each channel, and then:

[0037]

[0038] Further, in step 4, based on the strong correlation of the partial high frequency subbands, the grating templates are superimposed into the high frequency subbands, and the specific process is as follows:

[0039] Subband 1 and subband 6 have strong correlation, and subband 3 and subband 4 have strong correlation. Therefore, each channel can embed 4 grating templates, that is, subband 1 and 6 share one grating template, subband 3 and 4 share one grating template, and subband 2 and subband 5 each embed one grating template. Corresponding to one host image, k = 4 x 3 = 12 grating templates can be embedded. The embedding process is as follows:

[0040]

[0041] Wherein a is the embedding strength, and the value of b is determined by the correlation parameter d of the subband:

[0042]

[0043] Further, in step 5, the inverse DTCWT transform of the level is performed to obtain the RGB channel containing the watermark, and finally the image containing the watermark is obtained.

[0044] Further, in step A, the perspective transformation is performed on the image after shooting to obtain the corrected image.

[0045] Further, in step B, the corrected image is decomposed into RGB three channels, and the DTCWT transform of level 1 is performed on each channel to obtain a series of high frequency subbands containing angle grating templates.

[0046] Further, in step C, the high frequency subbands containing angle grating templates are preprocessed, that is, local histogram equalization is performed, and then the grating template signal is enhanced, so as to enhance the angle feature, and the specific process is as follows:

[0047] For the image after shooting and perspective transformation It is divided into RGB three channels, and the DTCWT transform is performed on each channel, and then the following is obtained:

[0048]

[0049] Then each is divided into non-overlapping 16 x 16 image blocks, and the histogram equalization operation is performed on each image block, so as to enhance the grating texture, and then highlight the angle feature of the grating.

[0050] ​​Further, in step D, for the pre-processed high-frequency subband, angle-based decoding is performed to obtain the extracted watermark information. The specific process is as follows:

[0051] (1) Mean processing of highly correlated subbands: since the highly correlated subbands embed the same grating template, after highlighting the angle features, the mean processing is performed on the subbands embedding the same grating template:

[0052]

[0053] and then the DTCWT high-frequency subband set containing the angle grating to be decoded is obtained

[0054] (2) Angle decoding based on the Fourier domain: different angle grating templates usually present peaks in a specific frequency band in the Fourier domain

[0055] The frequency band is located in an elliptical region:

[0056]

[0057] where f represents the frequency of the grating template, r and c are the height and width of the DTCWT high-frequency subband respectively, and u and v are the Cartesian coordinates in the Fourier domain.

[0058] The angle of the grating template is consistent with the angle formed by the coordinate (v max ,u max ) of the maximum value of the ellipse and the negative v-axis. Based on this

[0059] characteristic, the angle can be calculated based on the following formula:

[0060]

[0061] where is the extracted angle information, each i represents an encoded bit sequence, and the extracted watermark information w * is obtained by combining the encoding table consistent with the embedding end and the Gray code decoding. Subsequently, the BCH decoding of w * can obtain the extracted message m * .

[0062] The application proposes a robust image watermarking method based on the moire pattern to resist screen shooting, which makes the carrier present a moire pattern in the spatial domain by using the embedding mode of superimposing the high-frequency subband in the DTCWT domain and the grating template with different angles. In the extraction stage, the angle-based decoding method is adopted, and the blind extraction of the message sequence can be accurately completed without relying on the original image or the original message sequence. ​

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] Distortion during screen capture isn't ignored; it's considered a possible form of watermarking. Specifically, the present invention designs a watermarking scheme in which the watermark itself manifests as a "Moore pattern." Comprehensive experiments conducted at varying shooting distances, angles, lighting conditions, and with various capture and display devices demonstrate robustness in complex scenarios, enabling accurate extraction of embedded message sequences. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 This is a flowchart of an overall process of an embodiment of the present invention.

[0066] Figure 2 Schematic diagram of mapping Gray code to angle template according to an embodiment of the present invention.

[0067] Figure 3 Graph showing the correspondence between the grating template spatial domain and the Fourier domain according to an embodiment of the present invention.

[0068] Figure 4 Schematic diagram of the experimental arrangement of an embodiment of the present invention.

[0069] Figure 5 The following is a comparison table of PSNR between the embodiment of the present invention and the existing methods on the public data set.

[0070] Figure 6 The figure is a subjective visual comparison of the embodiment of the present invention and the existing methods on the public dataset.

[0071] Figure 7 The following is a comparison table of BER of the embodiment of the present invention and the existing methods against different shooting distances on a public dataset.

[0072] Figure 8 The following is a comparison table of BER of the embodiment of the present invention and the existing methods against different shooting angles on a public dataset. DETAILED DESCRIPTION

[0073] A robust image watermarking method based on moiré pattern to resist screen shooting

[0074] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0075] like Figure 1 As shown, the present invention implements a robust image watermarking method based on moiré patterns to resist screen capture. The method mainly includes a watermark embedding process and a watermark extraction process. The watermark embedding process includes watermark preprocessing, angle-based encoding, and watermark embedding; the watermark extraction process includes preprocessing and angle-based decoding.

[0076] The embodiment includes the following embedding steps:

[0077] S1: encode the watermark information into different angle raster templates, Figure 2 The mapping process from Gray code to angle raster template is given.

[0078] The steps of watermark preprocessing:

[0079] S1.1: for a message sequence m with length l m , encode it by BCH to get a sequence w with length l w . Then w is encoded by n-bit Gray code to get the final Gray coded sequence where s i is composed of n bits, and k = l w / n. In this embodiment, l m = 24, l w = 48, k = 12, and n = 4.

[0080] The steps of angle raster template mapping:

[0081] S1.2: construct a mapping function p from Gray code to angle:

[0082]

[0083] The formula maps the n-bit Gray code sequence where each g i is composed of n bits, to the corresponding angle set Specifically, this means dividing 180° into 2 n angle intervals, and each Gray code g i is mapped to the central angle θ i of the angle interval it belongs to:

[0084]

[0085] where Δθ represents the size of each angle interval, defined as:

[0086] S1.3: construct a mapping function q from angle to raster template:

[0087]

[0088] The above formula maps to the raster template where l i can be represented as:

[0089] l i(x, y) = p(φ i (x, y)),

[0090] p(u) = 0.5 + 0.5cos(2πf·u),

[0091] φ i (x, y) = xcosθ i +ysinθ i .

[0092] Here f represents the frequency of the grating, and the composite mapping can be obtained as follows:

[0093]

[0094] Based on the composite mapping, the Gray code sequence can be mapped to the set of grating templates In this embodiment, f = 1 / 6 and n = 4.

[0095] S2: The host image I is decomposed into RGB three channels to obtain I R , I G , and I B , and L-level DTCWT decomposition is performed, and 6 high-frequency signals (H) and one low-frequency signal can be obtained for each channel, and then the following is obtained:

[0096]

[0097] S3: Since subband 1 and subband 6 have strong correlation, and subband 3 and subband 4 have strong correlation, each channel can embed 4 grating templates, that is, subband 1 and 6 share one grating template, subband 3 and 4 share one grating template, and subband 2 and subband 5 each embed one grating template. Corresponding to one host image, k = 4 x 3 = 12 grating templates can be embedded. The embedding process is as follows:

[0098]

[0099] Wherein a is the embedding strength, and the value of b is determined by the correlation parameter d of the subband:

[0100]

[0101] In this embodiment, the embedding strength a = 0.2.

[0102] S4: Perform inverse DTCWT transformation of L level to obtain the RGB channel containing the watermark, and finally obtain the image containing the watermark.

[0103] This embodiment includes the following extraction steps:

[0104] ​SA: Perform perspective transformation on the captured image to obtain the corrected image

[0105] SB: Decompose the corrected image into RGB channels and perform The DT CWT transform of the level is used to obtain a series of high-frequency sub-bands containing angle grating templates.

[0106] SC: Preprocess the high-frequency sub-band containing the angle grating template to enhance the grating template signal and thus enhance the angle feature. The specific process is as follows:

[0107] For the image after shooting and perspective transformation Split it into RGB channels and perform DT CWT transformation on each channel to obtain:

[0108]

[0109] Then each The image is divided into non-overlapping 16×16 blocks, and a histogram equalization operation is performed on each image block to enhance the grating texture and highlight the angular characteristics of the grating.

[0110] SD: For the preprocessed high-frequency sub-band, angle-based decoding is performed to obtain the extracted watermark information.

[0111] SD.A performs mean processing on highly correlated sub-bands: Since highly correlated sub-bands are embedded in the same grating template, After the angle features are obtained, the sub-bands with the same embedding are averaged:

[0112]

[0113] Then we get the DT CWT high frequency subband set containing angle grating to be decoded

[0114] Grating templates at different SD.B angles usually show peaks in specific frequency bands in the Fourier domain, such as Figure 3 As shown. The frequency band is located in an elliptical area:

[0115]

[0116] where f represents the frequency of the grating template, r and c are the height and width of the DT CWT high-frequency subband, respectively, and u and v are the Cartesian coordinates in the Fourier domain.

[0117] The angle between the raster template and the coordinates of the maximum value of the ellipse (v max ,u max) is consistent with the angle formed by the negative v-axis. Based on this characteristic, the angle can be calculated using the following formula:

[0118]

[0119] in To extract the angle information, each θ i Represents an encoded bit sequence, combined with the encoding table consistent with the embedding end and Gray code decoding to obtain the extracted watermark information w * . Then, w * Perform BCH decoding to get the extracted message m * In this embodiment, f=1 / 6.

[0120] The payload of this embodiment is 24 bits, and BCH (24, 48) encoding is used to form a 48-bit codeword. When the number of error bits is ≤ 2, the message sequence can be completely extracted and verified. The test data is 100 images with a resolution of 512×512. In addition, the present invention uses a 24-inch display that supports a refresh rate of 50 / 60 / 75Hz, and a camera with a pixel of 4000w. A protractor, a ruler, lighting equipment and a tripod were also used during the shooting process. The layout of the specific experimental equipment is as follows: Figure 4 shown.

[0121] This example uses Peak Signal-to-Noise Ratio (PSNR) and Bit Error Rate (BER) as evaluation metrics: a higher PSNR value corresponds to better visual quality, and a lower BER value corresponds to stronger robustness.

[0122] This example is compared with three existing methods, which are: [1] TERA: Screen-to-camera image code with transparency, efficiency, robustness and adaptability, published in IEEE Transactions on Multimedia, Vol. 24, pp. 955-967, 2021; [2] Stegastamp: Invisible hyperlinks in physical photographs, published in Proceedings of the IEEE / CVF conference on computer vision and pattern recognition, pp. 2117-2126, 2020; and [3] Screen watermarking for data theft investigation and attribution, published in 10th International Conference on Cyber Conflict (CyCon), pp. 391-408, 2018. S Objective visual quality (PSNR) comparison results are shown, and the higher the PSNR value, the better the corresponding visual quality. The results show that the present application has good visual quality.

[0123] Figure 5 Objective visual quality (PSNR) comparison results are shown, and the higher the PSNR value, the better the corresponding visual quality. The results show that the present application has good visual quality. Figure 6 Subjective visual quality comparison is shown, and [1], [2] and [3] in the figure are the methods of the above prior art documents [1], [2] and [3], respectively. The results show that schemes [1] and [2] choose to represent watermark traces in dot matrix form, and thus the watermark artifacts are more obvious. Scheme [3] uses a deep learning method to fit the noise in the screen-to-camera, which usually causes serious visual distortion to the carrier. The present application displays watermark information based on the moire pattern, and the presented watermark artifacts are shallow, showing good characteristics in subjective visual quality. Figure 7 Comparison of the present application with existing methods at different shooting distances in the public data set, Figure 8 Comparison of the present application with existing methods at different shooting angles in the public data set. The results show that the present application has a low BER value under different shooting conditions.

Claims

1. A robust image watermarking method based on moiré patterns to achieve anti-screening protection. The watermark information is encoded in the designed moiré fringes of the image. By embedding the high-frequency subbands in the DT CWT domain with grating templates at different angles, the carrier presents a "moiré pattern" in the spatial domain. The method is characterized by: The embedding process and the extraction process comprise the following steps; The embedding process thereof comprises the following steps: Step 1: generating an encoded watermark sequence through watermark preprocessing of an original message sequence; Step 2: obtaining different-angle raster templates based on angle coding of the encoded watermark sequence in step 1; Step 3: For the original image, read its RGB channels and for each channel a series of high frequency subbands by a dual tree complex wavelet transform (DTCWT) transform. Step 4: superimposing the raster templates into the high-frequency subband based on the strong correlation of part of the high-frequency subband; Step 5: Perform inverse DTCWT transform of the high frequency subband after superimposing the grating mask obtained in step 4 to obtain the watermarked RGB channel, and then merge the three channels to obtain the watermarked image. Step 5: Perform inverse DTCWT transform of the high frequency subband after superimposing the grating mask obtained in step 4 to obtain the watermarked RGB channel, and then merge the three channels to obtain the watermarked image. The extraction process thereof comprises the following steps: Step A: performing perspective transformation on the photographed image to obtain a corrected image; Step B: The corrected image is decomposed into RGB three channels, and the DTCWT transform of level 1 is performed on each channel to obtain a series of high-frequency subbands containing angular grating templates. Step B: The corrected image is decomposed into RGB three channels, and the DTCWT transform of level 1 is performed on each channel to obtain a series of high-frequency subbands containing angular grating templates. Step C: pre-processing the high-frequency subband containing the angle raster template, that is, performing local histogram equalization to enhance the raster template signal and the angle feature; Step D: performing angle-based decoding on the pre-processed high-frequency subband to obtain the extracted watermark information; In step 2, the encoded watermark sequence is angle-coded to obtain different-angle raster templates, and the specific process is as follows: (1) constructing a mapping function p from the Gray code to the angle: The formula will generate an n-bit Gray code sequence where each g i is an n-bit Gray code sequence n consisting of n bits, mapped to a corresponding set of angles Specifically, 180° is divided into 2 i angle intervals, and each Gray code g i is mapped to the center angle θ of the corresponding angle interval. where Δθ represents the size of each angular interval, defined as: (2) constructing a mapping function q from the angle to the raster template: The above equation maps to the raster template where l i may be expressed as: l i (x,y) = p(φ i (x,y)), p(u) = 0.5 + 0.5cos(2πf·u), φ i (x,y) = x cos θ i + y sin θ i . where f represents the frequency of the raster, and the composite mapping can be obtained as follows: Based on the composite mapping, the Gray code sequence can be mapped to the set of raster templates In step 4, the raster template is embedded in the high-frequency subband of its DTCWT, and the specific process is as follows: Based on the correlation of the high-frequency subband of the DTCWT, subband 1 and subband 6 have strong correlation, and subband 3 and subband 4 have strong correlation, and four raster templates are embedded in each channel, that is, subband 1 and 6 share one raster template, subband 3 and 4 share one raster template, and subband 2 and subband 5 each embed one raster template; Corresponding to one host image, k = 4 × 3 = 12 raster templates are embedded; The embedding process is as follows: where Δ is the embedding strength, and the value of b is determined by the correlation parameter d of the subband: In step C, the high-frequency subband containing the angle raster template is pre-processed to enhance the raster template signal and the angle feature, and the specific process is as follows: For the image after shooting and after perspective transformation It is split into RGB three channels, and DTCWT transformation is performed on each channel, and then the following is obtained: Each of the images is then divided into non-overlapping 16x16 image blocks and a histogram equalization operation is performed on each image block to enhance the raster texture and thereby highlight the angular features of the raster. Each of the images is then divided into non-overlapping 16x16 image blocks and a histogram equalization operation is performed on each image block to enhance the raster texture and thereby highlight the angular features of the raster. In step D, for the pre-processed high-frequency subband, angle-based decoding is performed, and the specific process is as follows: (1) Mean processing on highly correlated subbands: Since highly correlated subbands embed the same grating template, mean processing on the subbands that embed the same is performed after highlighting the angular features: ​ Further, a set of DTCWT high frequency subbands of the angle raster containing image to be decoded is obtained (2) Angle decoding based on the Fourier domain: different-angle raster templates usually present peaks in a specific frequency band in the Fourier domain, which is located in an elliptical region: where f represents the frequency of the raster template, r and c are the height and width of the high-frequency subband of the DTCWT respectively, and u and v are the Cartesian coordinates in the Fourier domain; The angle of the grating template is consistent with the angle formed by the coordinate (v max ,u max ) of the maximum value of the ellipse and the negative v-axis, and the angle is calculated based on the following formula: wherein for the extracted angle information, each θ i represents an encoded bit sequence, and the extracted watermark information w is obtained in combination with an encoding table and a Golay code decoding consistent with the embedding end * ; then, the BCH decoding is performed on w * to obtain the extracted message m * .

2. A robust image watermarking method for anti-screen capture based on Moire pattern according to claim 1, characterized in that, In step 1, the process of generating an encoded watermark sequence through watermark preprocessing is as follows: For a message sequence m of length l m , it is BCH encoded to get a sequence w of length l w ; then w is n-bit Gray coded to get the final Gray coded sequence where s i is composed of n bits, k = l w / n.

3. A robust image watermarking method based on Moire pattern to achieve anti-screen capture as claimed in claim 1, wherein, In step 3, the RGB channel of the image is transformed by DTCWT, and the specific process is as follows: The host image I is decomposed into RGB three channels to obtain I R ,I G ,I B In the middle, and carry out Level DTCWT decomposition, each channel obtains 6 high frequency signals (H) and a low frequency signal, and further obtains:

4. The robust image watermarking method against screen capture based on Moire pattern of claim 1 or 2, wherein, The length of the Gray code is n = 4.

5. A robust image watermarking method based on Moire pattern to achieve anti-screen capture as claimed in claim 2, wherein, The message sequence m length l m = 24, the encoded watermark sequence w length l w = 48.

Images