A high capacity robust video watermarking method against geometric attacks

Abstract

This invention discloses a high-capacity robust video watermarking method resistant to geometric attacks, comprising: acquiring the YUV three channels of a video; applying DTCWT transform to the U channel to separate the coefficients in each sub-band; selecting a watermark embedding strategy, dividing the coefficients into blocks and applying DCT transform, modifying the DCT coefficients according to the watermark information to be embedded, and sequentially applying inverse DCT transform and inverse DTCWT transform to obtain the watermarked U channel; combining it with the Y and V channels of the original video and performing the inverse YUV conversion operation, finally obtaining the watermarked video through video encoding. The extraction process is based on DCT transform, and the watermark is extracted according to the embedding strategy. This invention achieves robust watermark embedding and extraction based on DTCWT and DCT, flexibly balancing the robustness of the watermark against geometric attacks with the capacity of the payload, and can be applied to multimedia platforms where video is the mainstream.

Description

Technical Field

[0001] This invention relates to the fields of multimedia technology and information security, specifically to a high-capacity robust video watermarking method resistant to geometric attacks, suitable for application scenarios requiring high robustness and high embedding capacity. Background Technology

[0002] The rapid development of high-speed networks has made the dissemination of digital media content on the internet much easier. People can easily upload and share their digital products through personal websites or social networks. However, these products may be downloaded and illegally distributed without authorization, which seriously harms the interests of the owners. Digital watermarking is one of the main technologies developed to protect the copyright of digital media content.

[0003] It embeds owner information or fingerprints (i.e., watermarks) into digital media content. Extracting these fingerprints can identify the ownership of digital media content or track its distribution history. As a multimedia format integrating images and audio, video is one of the main types of digital media content. In practical applications, video is easily interfered with by attacks such as compression, noise, or scaling during transmission. Therefore, the robustness of video watermarking algorithms is crucial. Existing robust video watermarking schemes can be divided into: compression domain watermarking, spatial domain watermarking, transform domain watermarking, and deep learning-based video watermarking technology. Compression domain watermarking schemes refer to watermark embedding based on the encoding process of standard codecs (such as MPEG, H.264, or H.265). The related technologies are relatively mature, and the latest schemes have achieved a good level of robustness against recompression attacks, but these schemes are less robust against geometric attacks. Spatial domain watermarking schemes embed watermarks by directly modifying the pixel values ​​of video frames. These schemes are generally less resistant to geometric attacks such as rotation attacks. Deep learning-based watermarking algorithms build and train networks to locate appropriate watermark embedding positions. It requires a large amount of data and computing power to learn an effective model, where the input video is usually low-resolution; to improve robustness, various types of attacks need to be predicted and simulated in the network, and performance may be poor for attacks that cannot be simulated. Transform domain watermarking schemes convert spatial video content to the transform domain and achieve watermark embedding by modifying the transform domain coefficients; these schemes have good performance in resisting common attacks such as recompression attacks and geometric attacks, and are receiving increasing attention in the field of video watermarking.

[0004] However, existing transform domain schemes have low embedding capacity, which limits their practical application scenarios, and their robustness against geometric attacks has not yet reached its optimal level. Summary of the Invention

[0005] The purpose of this invention is to provide a high-capacity robust video watermarking method resistant to geometric attacks. This method overcomes the shortcomings of current transform domain schemes in terms of low embedding capacity and further improves robustness, thereby meeting current practical needs.

[0006] To achieve the above functions, this invention designs a high-capacity robust video watermarking method resistant to geometric attacks. For the original video to be watermarked, the following steps S1-S5 are performed to obtain the watermarked video:

[0007] Step S1: Decode the original video to be watermarked to obtain a video frame sequence. Perform YUV conversion on each video frame in the video frame sequence to obtain the Y, U and V channels of each video frame of the original video.

[0008] Step S2: For each video frame's U channel, use DTCWT transform to obtain each low-frequency sub-band on the U channel, and then separate the coefficients of each low-frequency sub-band.

[0009] Step S3: Set a preset resolution threshold, compare the width of the original video to be watermarked with the resolution threshold, and select the appropriate watermark embedding strategy based on the comparison result.

[0010] Step S4: According to the selected watermark embedding strategy, the coefficients on the low-frequency sub-band corresponding to the watermark embedding strategy are divided into blocks, and DCT transformation is applied to obtain a preset number of DCT blocks. The DC coefficients in the DCT blocks are modified according to the watermark information to be embedded, and inverse DCT transformation and inverse DTCWT transformation are applied in sequence to obtain the watermarked U channel.

[0011] Step S5: Based on the watermarked U channel obtained in step S4, combine it with the Y and V channels of the original video to be watermarked and then perform the inverse YUV conversion operation. Finally, through video decoding, the watermarked video is obtained.

[0012] As a preferred technical solution of the present invention, the specific steps of step S1 are as follows:

[0013] Step S1.1: Input the original video to be embedded with the watermark into the video codec for decoding to obtain the video frame sequence corresponding to the original video, and convert the watermark information to be embedded into binary data through number system conversion;

[0014] Step S1.2: Perform RGB to YUV conversion on the video frame sequence to obtain the Y, U and V channels of each video frame of the original video.

[0015] As a preferred technical solution of the present invention, the specific steps of step S2 are as follows:

[0016] Step S2.1: Apply i-level DTCWT transform to the U channel of each video frame to obtain six low-frequency sub-bands as follows:

[0017]

[0018] In the formula, dtcwt i This represents the i-level DTCWT transform, where U represents the U channel of the video frame. This represents the j-th low-frequency subband obtained by applying i-level DTCWT transform to the U channel;

[0019] Step S2.2: After the U channel of each video frame undergoes DTCWT transformation, the coefficients of each low-frequency sub-band are separated to obtain the real and imaginary parts of the low-frequency sub-band respectively.

[0020]

[0021]

[0022] In the formula, real and img represent functions for extracting the real and imaginary digits from negative numbers, respectively. Indicates low-frequency subband The real part, Indicates low-frequency subband The imaginary part.

[0023] As a preferred technical solution of the present invention, the watermark embedding strategy in step S3 is specifically as follows:

[0024] If the width of the original video to be watermarked is greater than the resolution threshold, the highest level DTCWT transform is selected to separate the coefficients of each video frame of the original video to be watermarked. If the width of the original video to be watermarked is less than or equal to the resolution threshold, the highest level and the second highest level DTCWT transform are selected to separate the coefficients of each video frame of the original video to be watermarked.

[0025] As a preferred embodiment of the present invention, the specific steps of step S4 are as follows:

[0026] Step S4.1: Based on the watermark embedding strategy selected in step S3, select the corresponding level of DTCWT transform to perform coefficient separation on each video frame of the original video to be watermarked, and obtain the real part of the low-frequency subband of the corresponding level. Divide it into N non-overlapping 8×8 real number blocks, and apply DCT transformation to each real number block to obtain DCT blocks. The block division and DCT transform are as follows:

[0027]

[0028]

[0029] In the formula, This represents the m-th real block in the j-th low-frequency subband at the i-level. Representing blocks of real numbers The DCT block obtained after DCT transformation, where Block represents the block function and dct represents the DCT transformation;

[0030] Step S4.2: Based on the binary watermark information w to be embedded, modify the DC coefficient coff in the DCT block to coff. w The specific formula is as follows:

[0031]

[0032] In the formula, This represents the coefficient located at position [,1,1] in the real number block;

[0033] When the binary watermark information w=1, the formula for modifying the coefficient to embed the watermark is as follows:

[0034]

[0035] When the binary watermark information w = 0, the formula for modifying the coefficient to embed the watermark is as follows:

[0036]

[0037] In the formula, T is the robustness threshold;

[0038] Step S4.3: For the watermarked DCT block obtained in step S4.2, apply inverse DCT transform, and combine it with the imaginary part of the low-frequency sub-band obtained in step S2.2, apply inverse DTCWT transform to obtain the watermarked U channel U. w .

[0039] As a preferred embodiment of the present invention, the specific steps of step S5 are as follows:

[0040] Step S5.1: Convert the watermarked U channel obtained in step S4.3 into U w Combine the Y and V channels of the video frames of the original video obtained in step S1, and perform YUV to RGB conversion to obtain a video frame sequence with watermark.

[0041] Step S5.2: Input the watermarked video frame sequence into the video codec to obtain the watermarked video.

[0042] As a preferred technical solution of the present invention: for videos containing watermarks, the method further includes a process for extracting the watermark from the video, specifically including the following steps S6-S7:

[0043] Step S6: For the watermarked video, use the same method as in Step S1 to obtain the Y, U, and V channels of each video frame of the watermarked video; use the same method as in Step S2 to obtain the watermark coefficient coff. w ′ ;

[0044] Step S7: Extract the watermark sequence according to the embedding strategy. The extraction formula is as follows:

[0045]

[0046] In the formula, W = {W i}, where W is the watermark sequence.

[0047] Beneficial effects: Compared with the prior art, the advantages of the present invention include:

[0048] (1) This invention proposes a high-capacity robust video watermark that resists geometric attacks. The robustness of the watermarked video against common noise and geometric attacks can be basically achieved without loss. At the same time, it proposes a watermark embedding strategy that is adaptively selected according to the video resolution, which flexibly balances the robustness of the watermark against attacks and the capacity of the payload, and can be applied to multimedia platforms where video is the mainstream.

[0049] (2) The peak signal-to-noise ratio between the watermarked video generated by this invention and the original video can reach up to 52.82dB, and the average value can reach 46.51dB, which is higher than the existing transform domain robust video watermarking methods. Attached Figure Description

[0050] Figure 1 This is a framework diagram of a high-capacity robust video watermarking method against geometric attacks provided by an embodiment of the present invention;

[0051] Figure 2 This is a flowchart of watermark embedding provided according to an embodiment of the present invention;

[0052] Figure 3 This is a flowchart of watermark extraction provided according to an embodiment of the present invention. Detailed Implementation

[0053] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0054] Reference Figure 1This invention provides a high-capacity robust video watermarking method resistant to geometric attacks, including a watermark embedding process and a watermark extraction process. The copyright owner performs the watermark embedding process, and the specific method is as follows: Refer to... Figure 2 For the original video to be watermarked, perform the following steps S1-S5 to obtain the video with watermark;

[0055] Step S1: Decode the original video to be watermarked to obtain a video frame sequence. Perform YUV conversion on each video frame in the video frame sequence to obtain the Y, U and V channels of each video frame of the original video.

[0056] The image data used in this embodiment comes from the YouTube-VOS HD dataset and the CIF dataset commonly used in video encoders. In implementation, this embodiment uses MATLAB for programming and attack simulation experiments, and FFMPEG as the video codec.

[0057] The specific steps of step S1 are as follows:

[0058] Step S1.1: Input the original video to be embedded with the watermark into the video codec for decoding to obtain the video frame sequence corresponding to the original video, and convert the watermark information to be embedded into binary data through number system conversion;

[0059] In this embodiment, different watermark information carriers, such as text or icons, are converted into binary data for watermark embedding. Similarly, after extracting the binary data, it must be converted back to the original watermark information format. Existing transform domain schemes have a single-frame effective payload of only 1 bit, making it difficult to embed information such as text or icons.

[0060] Step S1.2: Perform RGB to YUV conversion on the video frame sequence to obtain the Y, U and V channels of each video frame of the original video.

[0061] In this embodiment, the U channel is selected to embed the watermark. Since the R, G, and B channels of a video frame are equally important, and modifying any one of them would significantly impact the visual quality of the video frame, they are moved to the YUV channel. To achieve a good balance between robustness and imperceptibility, the U channel, which has the least impact, is chosen as the watermark carrier.

[0062] Step S2: For the U channel of each video frame, use DTCWT (Dual-Tree Complex Wavelet Transform) to obtain each low-frequency sub-band on the U channel, and separate the coefficients of each low-frequency sub-band.

[0063] The specific steps of step S2 are as follows:

[0064] Step S2.1: Apply i-level DTCWT transform to the U channel of each video frame to obtain six low-frequency sub-bands as follows:

[0065]

[0066] In the formula, dtcwt i This represents the i-level DTCWT transform, where U represents the U channel of the video frame. This represents the j-th low-frequency subband obtained by applying i-level DTCWT transform to the U channel;

[0067] Step S2.2: After the U channel of each video frame undergoes DTCWT transformation, coefficient separation is performed on each low-frequency sub-band. Coefficient separation involves separating the real and imaginary parts of the low-frequency sub-band to obtain the real and imaginary parts of the low-frequency sub-band respectively.

[0068]

[0069]

[0070] In the formula, real and img represent functions for extracting the real and imaginary digits from negative numbers, respectively. Indicates low-frequency subband The real part, Indicates low-frequency subband The imaginary part.

[0071] In this embodiment, based on the complex nature of the DTCWT coefficients, the real part of the DTCWT coefficients is selected as the carrier, reducing the visual impact of modifying the entire coefficient and thus improving the visual quality of the watermarked video. Furthermore, instead of selecting only one or two pairs of sub-bands to embed redundant watermark information, this embodiment embeds different watermark information in each of the six low-frequency sub-bands, thereby increasing the embedding capacity.

[0072] Step S3: Set a preset resolution threshold, compare the width of the original video to be watermarked with the resolution threshold, and select the appropriate watermark embedding strategy based on the comparison result.

[0073] The watermark embedding strategy described in step S3 is as follows:

[0074] In this embodiment, the resolution threshold is set to 500. If the width of the original video to be watermarked is greater than the resolution threshold, the highest level DTCWT transform is selected to separate the coefficients of each video frame of the original video to be watermarked. If the width of the original video to be watermarked is less than or equal to the resolution threshold, the highest level and the second highest level DTCWT transform are selected simultaneously to separate the coefficients of each video frame of the original video to be watermarked.

[0075] When low video resolution results in low embeddable capacity, a strategy of expanding the embedding area is adopted to increase the embedding capacity of the watermark; while when the video resolution is high, a strategy of choosing the highest level as the watermark carrier is adopted to maintain robustness and visual quality when the embedding capacity space is sufficient.

[0076] Step S4: According to the selected watermark embedding strategy, the coefficients on the low-frequency sub-band corresponding to the watermark embedding strategy are divided into blocks, and DCT (Discrete Cosine Transform) is used to obtain a preset number of DCT blocks. The DC coefficients in the DCT blocks are modified according to the watermark information to be embedded, and the inverse DCT and inverse DTCWT transformations are applied in sequence to obtain the watermarked U channel.

[0077] The specific steps of step S4 are as follows:

[0078] Step S4.1: Based on the watermark embedding strategy selected in step S3, select the corresponding level of DTCWT transform to perform coefficient separation on each video frame of the original video to be watermarked, and obtain the real part of the low-frequency subband of the corresponding level. Divide it into N non-overlapping 8×8 real number blocks, and apply DCT transformation to each real number block to obtain DCT blocks. The block division and DCT transform are as follows:

[0079]

[0080]

[0081] In the formula, This represents the m-th real block in the j-th low-frequency subband at the i-level. Representing blocks of real numbers The DCT block obtained after DCT transformation, where Block represents the block function and dct represents the DCT transformation;

[0082] In this embodiment, the size is set to 8 to improve the watermark embedding capacity when performing block operations on real numbers, based on the video encoding process. A size of 4 results in too small a block granularity and low robustness against attacks in preserving information; a size of 16 results in too large a block granularity, which would reduce the effective payload capacity in this embodiment. Therefore, setting the size to 8 is more appropriate.

[0083] Step S4.2: Based on the binary watermark information w to be embedded, modify the DC coefficient coff in the DCT block to coff. w The specific formula is as follows:

[0084]

[0085] In the formula, This represents the coefficient at position [1,1] in the real number block, i.e., the DC coefficient;

[0086] When the binary watermark information w=1, the formula for modifying the coefficient to embed the watermark is as follows:

[0087]

[0088] When the binary watermark information w = 0, the formula for modifying the coefficient to embed the watermark is as follows:

[0089]

[0090] In the formula, T is the robustness threshold. The larger T is, the stronger the robustness, but the greater the impact on the visual quality of the video. Therefore, this embodiment tested the visual quality and robustness of watermarked videos by setting different T values, and finally selected 15 as the threshold T value to achieve a better balance between robustness and visual quality. It should be noted that in this embodiment, only the coefficients with values ​​in the range [-T, T] are modified, and those outside this range are not modified to minimize the impact on visual quality.

[0091] Step S4.3: For the watermarked DCT block obtained in step S4.2, apply inverse DCT transform, and combine it with the imaginary part of the low-frequency sub-band obtained in step S2.2, apply inverse DTCWT transform to obtain the watermarked U channel U. w .

[0092] In this embodiment, only the real part is modified because the robustness of the real part is higher than that of the imaginary part. Also, considering the visual quality of the watermarked video, the imaginary part of the DTCWT is not modified.

[0093] Step S5: Based on the watermarked U channel obtained in step S4, combine it with the Y and V channels of the original video to be watermarked and then perform the inverse YUV conversion operation. Finally, through video decoding, the watermarked video is obtained.

[0094] The specific steps of step S5 are as follows:

[0095] Step S5.1: Convert the watermarked U channel obtained in step S4.3 into U w Combine the Y and V channels of the video frames of the original video obtained in step S1, and perform YUV to RGB conversion to obtain a video frame sequence with watermark.

[0096] Step S5.2: Input the watermarked video frame sequence into the video codec to obtain the watermarked video.

[0097] The copyright verifier performs the watermark extraction process, referring to... Figure 3 The specific method is as follows: For videos with watermarks, it also includes the process of extracting the watermark from the video, specifically including the following steps S6-S7;

[0098] Step S6: For the watermarked video, use the same method as in Step S1 to obtain the Y, U, and V channels of each video frame of the watermarked video; use the same method as in Step S2 to obtain the watermark coefficient coff. w ′ ;

[0099] Step S7: Extract the watermark sequence according to the embedding strategy. The extraction formula is as follows:

[0100]

[0101] In the formula, W = {W i}, where W is the watermark sequence.

[0102] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A high-capacity robust video watermarking method resistant to geometric attacks, characterized in that, For the original video to be watermarked, perform the following steps S1-S5 to obtain the watermarked video: Step S1: Decode the original video to be watermarked to obtain a video frame sequence. Perform YUV conversion on each video frame in the video frame sequence to obtain the Y, U and V channels of each video frame of the original video. Step S2: For each video frame's U channel, use DTCWT transform to obtain each low-frequency sub-band on the U channel, and then separate the coefficients of each low-frequency sub-band. The specific steps of step S2 are as follows: Step S2.1: Apply i-level DTCWT transform to the U channel of each video frame to obtain six low-frequency sub-bands as follows: ； In the formula, This represents the i-level DTCWT transform, where U represents the U channel of the video frame. This represents the j-th low-frequency subband obtained by applying i-level DTCWT transform to the U channel; Step S2.2: After the U channel of each video frame undergoes DTCWT transformation, the coefficients of each low-frequency sub-band are separated to obtain the real and imaginary parts of the low-frequency sub-band respectively. ； ； In the formula, , These represent functions that extract the real and imaginary digits from negative numbers, respectively. Indicates low-frequency subband The real part, Indicates low-frequency subband The imaginary part; Step S3: Set a preset resolution threshold, compare the width of the original video to be watermarked with the resolution threshold, and select the appropriate watermark embedding strategy based on the comparison result. The watermark embedding strategy described in step S3 is as follows: If the width of the original video to be watermarked is greater than the resolution threshold, the highest level DTCWT transform is selected to separate the coefficients of each video frame of the original video to be watermarked. If the width of the original video to be watermarked is less than or equal to the resolution threshold, the highest level and the second highest level DTCWT transform are selected to separate the coefficients of each video frame of the original video to be watermarked. Step S4: According to the selected watermark embedding strategy, the coefficients on the low-frequency sub-band corresponding to the watermark embedding strategy are divided into blocks, and DCT transformation is applied to obtain a preset number of DCT blocks. The DC coefficients in the DCT blocks are modified according to the watermark information to be embedded, and inverse DCT transformation and inverse DTCWT transformation are applied in sequence to obtain the watermarked U channel. The specific steps of step S4 are as follows: Step S4.1: Based on the watermark embedding strategy selected in step S3, select the corresponding level of DTCWT transform to perform coefficient separation on each video frame of the original video to be watermarked, and obtain the real part of the low-frequency subband of the corresponding level. Divide it into N non-overlapping groups. The real number blocks are obtained, and for each real number block, the DCT transform is applied to obtain the DCT blocks. The block division and DCT transform are as follows: ； ； In the formula, This represents the m-th real block in the j-th low-frequency subband at the i-level. Representing blocks of real numbers The DCT block obtained by DCT transformation, Represents a block function. Indicates DCT transform; Step S4.2: Based on the binary watermark information w to be embedded, modify the DC coefficient coff in the DCT block to... The specific formula is as follows: ； In the formula, Indicates the position in the real number block as The coefficient; When the binary watermark information w=1, the formula for modifying the coefficient to embed the watermark is as follows: ； When the binary watermark information w=0, the formula for modifying the coefficient to embed the watermark is as follows: ； In the formula, T is the robustness threshold; Step S4.3: For the watermarked DCT block obtained in step S4.2, apply inverse DCT transform, and combine it with the imaginary part of the low-frequency sub-band obtained in step S2.2, apply inverse DTCWT transform to obtain the watermarked U channel. ; Step S5: Based on the watermarked U channel obtained in step S4, combine it with the Y and V channels of the original video to be watermarked and then perform the inverse YUV conversion operation. Finally, through video decoding, the watermarked video is obtained.

2. The high-capacity robust video watermarking method against geometric attacks according to claim 1, characterized in that, The specific steps of step S1 are as follows: Step S1.1: Input the original video to be embedded with the watermark into the video codec for decoding to obtain the video frame sequence corresponding to the original video, and convert the watermark information to be embedded into binary data through number system conversion; Step S1.2: Perform RGB to YUV conversion on the video frame sequence to obtain the Y, U and V channels of each video frame of the original video.

3. The high-capacity robust video watermarking method against geometric attacks according to claim 1, characterized in that, The specific steps of step S5 are as follows: Step S5.1: Convert the watermarked U channel obtained in step S4.3 Combine the Y and V channels of the video frames of the original video obtained in step S1, and perform YUV to RGB conversion to obtain a video frame sequence with watermark. Step S5.2: Input the watermarked video frame sequence into the video codec to obtain the watermarked video.

4. The high-capacity robust video watermarking method against geometric attacks according to claim 3, characterized in that, For videos containing watermarks, the process also includes extracting the watermark from the video, specifically including the following steps S6-S7: Step S6: For the watermarked video, use the same method as in Step S1 to obtain the Y, U, and V channels of each video frame of the watermarked video; use the same method as in Step S2 to obtain the watermark coefficients. ; Step S7: Extract the watermark sequence according to the embedding strategy. The extraction formula is as follows: ； In the formula, W is the watermark sequence.

Images