A feature point enhanced local watermark embedding and extraction method under perception constraint and guidance

Abstract

The application discloses a feature point enhanced local watermark embedding and extraction method under the perception of constraints and guidance. The embedding method comprises the following steps: SIFT feature point detection is performed on an image to be embedded with a watermark, and a plurality of feature points are obtained; the feature points are sorted according to the intensity of the feature points from large to small, and a plurality of non-overlapping regions positioned by a plurality of feature points with the highest intensity are selected as watermark embedding regions; each watermark embedding region is blocked according to the length of a watermark information bit; one bit of watermark is embedded in each block of each watermark embedding region by using a watermark embedding strategy of fused feature point enhancement; and the watermark embedding region after embedding the watermark is used to replace the corresponding region in the image to be embedded with the watermark, so that a watermark-containing image is obtained. Compared with the existing adaptive local image watermark embedding and extraction method, the application has better invisibility and stronger robustness.

Description

Technical Field

[0001] This invention relates to the field of digital image copyright protection, specifically to a method for feature point enhancement local watermark embedding and extraction under perceptual constraints and guidance. Background Technology

[0002] While the development of online media has brought convenience, it has also led to phenomena such as plagiarism and illegal dissemination of digital works, resulting in serious copyright disputes. Digital image watermarking technology conceals copyright information within images. In the event of copyright infringement, the watermark information can be extracted for copyright authentication and infringement tracking. Watermarking technology should possess both imperceptibility and robustness. The former requires that the embedded watermark does not affect the visual quality of the image, while the latter ensures that the watermark can be correctly extracted even after the image has undergone various attacks. Ensuring the imperceptibility of the watermark while maximizing its resistance to attacks is key to the effective application of watermarking technology.

[0003] Local watermarking embeds watermark information into multiple non-overlapping regions of an image. This local embedding reduces the visual impact on the overall image, improving the watermark's imperceptibility. Due to the multiple embedding, even if some regions are damaged, it won't affect the watermark in other regions, making it more robust. This technique typically uses feature points in the image, such as SIFT (Scale-Invariant Feature Transform), to locate the embedding region. Consistent feature points must be detected in both the watermark embedding and extraction stages to accurately locate the embedding region for successful watermark extraction. Attacks can alter image content, potentially causing feature point shifts or loss, making watermark extraction impossible. Some research attempts to enhance feature point stability by modifying pixel values ​​around them. However, since the watermark embedding process also requires modifying pixel values ​​within the same region, these two modifications may produce contradictory effects, reducing their effectiveness and affecting the watermark's imperceptibility and robustness.

[0004] Watermark embedding essentially involves making subtle modifications to image pixels according to a specific scheme. Generally, the larger the pixel modification, the more difficult the embedded watermark is to remove, and the stronger its robustness; however, it also has a greater impact on image quality, reducing the watermark's imperceptibility. Typically, a parameter is introduced into watermarking algorithms to control the overall pixel modification amount, balancing the watermark's imperceptibility and robustness. Some studies adaptively adjust the value of this parameter based on the image's statistical characteristics or content features, making the modification amount vary according to the image content, aiming to make the embedded watermark less noticeable. However, these methods can only adjust the overall modification amount of the image or region, not the modification amount of individual pixels. Because the human eye has different sensitivities to different pixels, the ability of different pixels to hide watermarks varies, and simply controlling the overall modification amount cannot ensure that changes to each pixel are imperceptible to the human eye. Summary of the Invention

[0005] To address the issue of poor watermark robustness caused by feature point offset after watermark embedding in local watermarking algorithms, and to overcome the shortcomings of existing watermarking technologies in optimizing watermark invisibility, this invention proposes a feature point-enhanced local watermark embedding and extraction method under perceptual constraints and guidance. This method maximizes the robustness of watermark extraction while ensuring the invisibility of the embedded watermark.

[0006] According to one aspect of the present invention, a feature point enhancement local watermark embedding method under perceptual constraints and guidance is provided, the method specifically comprising:

[0007] SIFT feature point detection is performed on the image to be embedded with watermark to obtain several feature points;

[0008] The feature points are sorted from largest to smallest intensity, and multiple non-overlapping regions located by the feature points with the highest intensity are selected as the watermark embedding regions.

[0009] Each watermark embedding region is divided into blocks according to the length of the watermark information bits;

[0010] In each block of each watermark embedding region, a one-bit watermark is embedded using a watermark embedding strategy enhanced by fusion feature points;

[0011] The corresponding area in the image to be embedded with watermark is replaced by the watermark embedding area after watermark embedding to obtain a watermarked image.

[0012] Wherein, embedding a one-bit watermark in each block of each watermark embedding region using a watermark embedding strategy enhanced by fused feature points includes:

[0013] Based on a pre-established quantization parameter model based on PSNR (Peak Signal to Noise Ratio), the embedding parameters in the fusion feature point enhanced watermark embedding strategy are determined.

[0014] The blocks are classified according to whether they contain feature points. Blocks containing feature points are called central image blocks, and blocks without feature points are called non-central image blocks.

[0015] Guided by JND (Just Noticeable Distortion), a one-bit watermark is embedded in each of the non-center image blocks using a quantization method. The quantization formula is shown in formula (1), the weights are modified as shown in formula (2), and the embedding formula is shown in formula (3).

[0016] (1)

[0017] (2)

[0018] (3)

[0019] in, i, j Indicates the index of the image patch. B i,j and B′ i,j These represent the non-central image blocks in the image to be watermarked and the watermarked image, respectively. i,j and u′ i,j They represent B respectively i,j and B′ i,j The average pixel value. q This is the embedding parameter, also known as the quantization step size, used to control the amount of pixel modification. i,j This indicates the watermark position to be embedded. `round()` is a rounding function, and `mod` represents the modulo operation. `Jnd` i,j It is image block B i,j The JND matrix for each pixel, M i,j This represents the pixel modification weight of the non-center image patch. `sum()` is the summation function.

[0020] Guided by JND, a one-bit watermark is embedded in the central image block using an optimization function. The weight matrix is ​​shown in Equation (4), and the optimization function is shown in Equation (5):

[0021] (4)

[0022] (5)

[0023] (6)

[0024] (7)

[0025] Among them, B o The watermark image I to be embedded contains feature points p. o The central image block, Jnd o B o The JND matrix. M o This represents the pixel modification weight of the central image block. I′ is the watermarked image, B o ′ is the watermarked image with B o The corresponding image patch. min represents taking the minimum value, ||M o (B o -B o ′)|| denotes the L2 norm. Constraint (C.1) guarantees p o The intensity increases to the specified value. o :maximum and p o :minimun respectively represent p o It is S centered on itself po The maximum and minimum intensity feature points within the neighborhood. (D) I (p o ) represents feature point p o The initial intensity value, D I’ (p o ) represents feature point p o The enhanced strength value This represents the enhancement value of the feature point intensity. Constraint (C.2) guarantees S po No new feature points were generated in the neighborhood. max and p min S represents po The maximum and minimum feature strengths within the neighborhood. Constraint (C.3) guarantees the embedding of the watermark.

[0026] The calculation method for the PSNR-based quantization parameter model specifically includes:

[0027] According to the embedding formula (1), assuming that each pixel in the block modifies Δu i,j The mean square error (MSE) of the entire watermark embedding region and the PSNR satisfy the following relationship.

[0028] (8)

[0029] Using Δu i,j 2 Expected value l q2 / 4 Estimate the MSE of the embedded region, substitute it into formula (8), and obtain the PSNR-based quantization parameter model as follows:

[0030] (9)

[0031] The value of the embedding parameter is calculated using the given PSNR value according to formula (9).

[0032] According to another aspect of the present invention, a method for extracting local watermarks with enhanced feature points under perceptual constraints and guidance is provided. This method is used to extract watermarks embedded using the perceptual constraint and guidance-based local watermark embedding method provided by the present invention, specifically including:

[0033] SIFT feature point detection is performed on the watermarked image to obtain several feature points;

[0034] The feature points are sorted from largest to smallest intensity, and multiple non-overlapping regions located by the feature points with the highest intensity are selected as the watermark extraction regions.

[0035] Each watermark extraction region is divided into blocks according to the length of the watermark information bits;

[0036] One watermark is extracted from each block of each watermark extraction region to obtain the complete watermark information. The extraction formula is as follows:

[0037] (10)

[0038] in, i, j Indicates the index of the image patch. This represents the calculated mean value of the image patch. This indicates the extracted watermark position. round() is a rounding function, and mod represents the modulo operation.

[0039] Compared with existing technologies, the above technical solution has at least the following beneficial effects:

[0040] This invention integrates feature point enhancement into the watermark embedding process. By modifying image pixels in a single step, it simultaneously achieves watermark embedding and feature point enhancement, avoiding mutual interference between the two processes and effectively improving the stability of feature points and the robustness of the watermark. A PSNR-based quantization parameter model is used to limit the total amount of pixel modification, and a JND-guided pixel-level weighting strategy is employed to differentiate the total modification amount across different pixels. This ensures good imperceptibility of the watermark at both the overall image block and individual image pixel levels. This invention demonstrates excellent robustness against most common image attacks and geometric attacks, effectively solving the problem of poor robustness caused by feature point offset in local image watermarking methods. Attached Figure Description

[0041] The accompanying drawings, as part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments and descriptions of the invention are used to explain the invention, but do not constitute an undue limitation of the invention. Obviously, the drawings described below are merely some embodiments, and those skilled in the art can obtain other drawings based on these drawings without creative effort. In the drawings:

[0042] Figure 1 This is a flowchart illustrating a perceptually constrained and guided feature point enhancement local watermarking embedding method according to an exemplary embodiment.

[0043] Figure 2 The image to be embedded with a watermark is shown according to an exemplary embodiment;

[0044] Figure 3 A watermarked image is shown according to an exemplary embodiment;

[0045] Figure 4 This illustrates the variation in the intensity of corresponding feature points in the image to be watermarked and the watermarked image, according to an exemplary embodiment.

[0046] Figure 5 This is a comparison of the PSNR values ​​of the embedding region in the image to be embedded with a watermark and the corresponding region in the watermarked image, according to an exemplary embodiment.

[0047] Figure 6 This is a flowchart illustrating a perceptually constrained and guided feature point enhanced local watermark extraction method according to an exemplary embodiment.

[0048] Figure 7 This is a histogram distribution of the mean of an image patch, as illustrated in an exemplary embodiment.

[0049] These accompanying drawings and textual descriptions are not intended to limit the scope of the invention in any way, but rather to illustrate the concept of the invention to those skilled in the art by reference to specific embodiments. Detailed Implementation

[0050] The technical problems solved by the embodiments of the present invention, the technical solutions adopted, and the technical effects achieved will be clearly and completely described below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other equivalent or obvious variations of embodiments obtained by those skilled in the art without creative effort fall within the protection scope of the present invention. The embodiments of the present invention can be embodied in various different ways as defined and covered by the claims.

[0051] It should be noted that many specific details are given in the following description for ease of understanding. However, it is obvious that the present invention may be implemented without these specific details.

[0052] It should be noted that, in the absence of explicit limitations or conflicts, the various embodiments and their technical features in this invention can be combined with each other to form a technical solution.

[0053] This invention provides a method for enhancing local watermarking by feature points under perceptual constraints and guidance. For example... Figure 1 As shown, the method may include steps S1000 to S1400.

[0054] Step S1000: Perform SIFT feature point detection on the image to be embedded with watermark to obtain several feature points.

[0055] The specific description is as follows: using different variances A Gaussian filter kernel is used to filter the image to be embedded with the watermark, resulting in a series of Gaussian filtered images. The difference between adjacent Gaussian filtered images is extracted to establish a Difference of Gaussian (DOG) image. This represents a position in the DOG (Domain of Image) located at row x, column y. D(p) represents the value of position p in the DOG image. ≤ 1, This represents 26 locations within a 3×3×3 cube centered at location p in the DOG image. and These represent the row and column coordinates of a pixel in a DOG image, respectively. Let D(p) be the variance value. Let D(p) be the variance of the 26-neighborhood S. pThe values ​​in the image are compared, and if D(p) is the maximum or minimum value, then position p is considered an extreme point in the image. Extreme points with low contrast and at edges are excluded to obtain SIFT feature points.

[0056] Step S1100: Sort the feature points according to their intensity from largest to smallest, and select multiple non-overlapping regions located by the multiple feature points with the highest intensity as the watermark embedding region.

[0057] The intensity of the feature points is defined as |D(p)|. The feature points are sorted from largest to smallest according to their intensity, and the non-overlapping region centered on the top N feature points is selected as the watermark embedding region. The ratio of the size of the watermark embedding region to the size of the image to be embedded with the watermark is K.

[0058] Step S1200: Divide each watermark embedding region into blocks according to the length L of the watermark information bits. Divide the watermark embedding region into a×b image blocks according to the watermark length L, ensuring that a and b are as equal as possible, where a and b are two positive integers and satisfy a×b≥L. Let the length of the image block be h and the width be w.

[0059] Step S1300: Embed a one-bit watermark in each block of each watermark embedding region using a watermark embedding strategy enhanced by fusion feature points.

[0060] Specifically, this step may include steps S1310 to S1340.

[0061] Step S1310: Determine the embedding parameters in the watermark embedding strategy enhanced by the fusion feature points according to the pre-established PSNR-based quantization parameter model.

[0062] The reason for performing this step is: to calculate the embedding parameter-constrained pixel modification amount Δu using the PSNR-based quantization parameter model. i,j This can keep the PSNR of the watermarked image consistent with the preset value, ensuring that the overall invisibility of the watermark meets expectations.

[0063] Step S1320: Classify the blocks according to whether they contain feature points. Blocks containing feature points are called central image blocks, and blocks without feature points are called non-central image blocks.

[0064] Step S1330: Under the guidance of JND, a one-bit watermark is embedded in each of the non-center image blocks using a quantization method. The quantization formula is shown in formula (1), the weight modification is shown in formula (2), and the embedding formula is shown in formula (3).

[0065] (1)

[0066] (2)

[0067] (3)

[0068] in, i, j Indicates the index of the image patch. B i,j and B′ i,j These represent the non-central image blocks in the image to be watermarked and the watermarked image, respectively. i,j and u′ i,j They represent B respectively i,j and B′ i,j The average pixel value. q This is the embedding parameter, also known as the quantization step size, used to control the amount of pixel modification. i,j This indicates the watermark position to be embedded. `round()` is a rounding function, and `mod` represents the modulo operation. `Jnd` i,j It is image block B i,j The JND matrix for each pixel, M i,j This represents the pixel modification weights of the non-center image patch. `sum()` is the summation function. `if` indicates an option.

[0069] The reason for performing the aforementioned steps is that the human eye has varying sensitivities to different pixels in the spatial domain, resulting in different watermark concealment capabilities among pixels. Simply controlling the overall modification amount cannot ensure that changes to each pixel are imperceptible to the human eye. It is necessary to further allocate the modification amount to different pixels based on their concealment capabilities. Therefore, to ensure that image modifications are consistent with human visual perception and to optimize watermark imperceptibility, this invention, under the guidance of JND, weights the quantized value modification amount (i.e., the total pixel modification amount) to each pixel when modifying pixels in the spatial domain, allocating more modification amount to pixels with lower human eye sensitivity (stronger concealment capabilities).

[0070] Step S1340: Guided by JND, embed a one-bit watermark in the central image block using an optimization function. The weight matrix is ​​shown in Equation (4), and the optimization function is shown in Equation (5):

[0071] (4)

[0072] (5)

[0073] (6)

[0074] (7)

[0075] Among them, B o The watermark image I to be embedded contains feature points p. oThe central image block, Jnd o B o The JND matrix. M o This represents the pixel modification weight of the central image block. I′ is the watermarked image, B o ′ is the watermarked image with B o The corresponding image patch. min represents taking the minimum value, ||M o (B o -B o ′)|| denotes the L2 norm. Constraint (C.1) guarantees p o The intensity increases to the specified value. o :maximum and p o :minimun respectively represent p o It is S centered on itself po The maximum and minimum intensity feature points within the neighborhood. (D) I (p o ) represents feature point p o The initial intensity value, D I’ (p o ) represents feature point p o The enhanced strength value This represents the enhancement value of the feature point intensity. Constraint (C.2) guarantees S po No new feature points were generated in the neighborhood. max and p min Representing the neighborhood S respectively po The maximum and minimum feature strengths within the watermark. Constraint (C.3) guarantees the embedding of the watermark.

[0076] The reason for performing the aforementioned steps is that if the region is modified twice independently, to achieve feature point enhancement and watermark embedding respectively, although the stability of the feature points can be improved, there are two main drawbacks: first, the feature point enhancement process introduces additional modifications, which reduces image quality; second, watermark embedding may weaken the feature point strength, causing the feature point enhancement to fail. Therefore, this invention achieves a watermark embedding algorithm that fuses feature point enhancement through a single modification, which can effectively enhance the stability of the feature points.

[0077] Specifically, the calculation method of the PSNR-based quantization parameter model in step S1310 may include steps S1311 to S1313.

[0078] Step S1311: According to the embedding formula (1), assume that the modification amount of each pixel in the block is... The mean square error (MSE) and PSNR of the entire watermark embedding region satisfy the following relationship:

[0079] (8)

[0080] The formula for calculating PSNR is as follows: The MSE is extracted through transformation and then combined with the MSE calculation formula. Formula (8) can be obtained.

[0081] Step S1312: Use Expected value l q 2 / 4 Estimate the MSE of the embedded region, substitute it into formula (8), and obtain the PSNR-based quantization parameter model as follows:

[0082] (9)

[0083] in, The calculation of the expected value is based on the following considerations:

[0084] During quantization, the amount of modification for each image patch is considered as a whole. All are distributed in [-l q, l q ]between, Distributed in [0 , l q The range is between ], and its value follows a mean of l. q A uniform distribution of / 2, from which we can obtain The expected value is l q 2 / 4.

[0085] To support the above conclusions, a preferred embodiment is provided: 24 images are randomly selected, each image is divided into 5 embedding regions, and each embedding region is further divided into 64 image blocks, for a total of 7680 image blocks. Assume l q =14, according to formula (1) we get 7680. The values ​​range from to , and their average value is 7.02. Their histogram distribution ( Figure 7 This can be confirmed. Follows the mean l q A uniform distribution of 2 / 2.

[0086] Step S1313: Calculate the value of the embedding parameter using the given PSNR value according to formula (9).

[0087] Step S1400: Replace the corresponding area in the image to be embedded with the watermark with the watermark-embedded area to obtain a watermarked image.

[0088] This invention provides a method for enhancing local watermark extraction using feature points under perceptual constraints and guidance. For example... Figure 6As shown, the method may include steps T1000 to T1300.

[0089] Step T1000: Using the same method as step S1000 of the feature point enhancement local watermark embedding method under perceptual constraints and guidance in Embodiment 1, SIFT feature point detection is performed on the watermarked image to obtain several feature points;

[0090] Step T1100: Using the same method as step S1100 of the feature point enhancement local watermark embedding method under perceptual constraints and guidance in Embodiment 1, the feature points are sorted from largest to smallest intensity, and multiple non-overlapping regions located by the multiple feature points with the highest intensity are selected as watermark extraction regions.

[0091] Step T1200: Using the same method as step S1200 of the feature point enhancement local watermark embedding method under perceptual constraints and guidance in Embodiment 1, each watermark extraction region is divided into blocks according to the length L of the watermark information bits;

[0092] Step T1300: Extract one watermark bit from each block of each watermark extraction region to obtain the complete L-bit watermark information. The extraction formula is as follows:

[0093] (10)

[0094] Where i and j represent the indices of the image patches. This represents the calculated mean value of the image patch. This indicates the extracted watermark position. round() is a rounding function, and mod represents the modulo operation.

[0095] like Figure 2 and 3 As shown, Figure 2 An image before watermarking is embedded is shown as an example, i.e., the image to be watermarked; Figure 3 An image with an embedded watermark is shown as an example, i.e., the watermarked image.

[0096] The following preferred embodiment will be used to describe in detail a feature point enhancement local watermark embedding method under perceptual constraints and guidance.

[0097] In this embodiment, one image is randomly selected as the image to be embedded with the watermark, such as... Figure 2 As shown, the image size is 512. 512.

[0098] This embodiment uses a 64-bit binary sequence as the original watermark. The length L of the watermark information is 64 bits, which can basically achieve a relatively comprehensive description of copyright information.

[0099] SIFT feature point detection was performed on the image to be embedded with the watermark, resulting in 100 feature points.

[0100] The feature points are sorted from highest to lowest intensity, and the five non-overlapping regions located by the five feature points with the highest intensity are selected as the watermark embedding regions. The ratio K between the size of the watermark embedding region and the size of the image to be embedded with the watermark is 1 / 6, which is 85. 85.

[0101] Each watermark embedding region is divided into 8×8 image blocks based on the length L = 64 of the watermark information bits.

[0102] The image blocks are classified according to whether they contain feature points. Image blocks containing feature points are called central image blocks, and image blocks without feature points are called non-central image blocks. Thus, for each watermark embedding region, one central image block and 63 non-central image blocks can be obtained.

[0103] The target PSNR is set to 40dB, and the quantization step size l is determined using the PSNR-based quantization parameter model formula (9). q The value is 15.

[0104] For the 63 non-center image blocks, the quantization watermark embedding shown in formulas (1), (2) and (3) is performed respectively.

[0105] For the single central image block, watermark embedding based on the optimization function is performed as shown in formulas (4), (5), (6), and (7). The optimization function is used to find pixel modification schemes that can simultaneously embed the watermark and enhance the intensity of feature points. The enhancement value of the embedding intensity is... Set it to 3.

[0106] Find all the central and non-central image blocks that meet the conditions, replace the corresponding image blocks in the image to be embedded with the watermark, realize the watermark embedding with enhanced fusion feature points, and obtain the watermarked image.

[0107] It should be noted that:

[0108] If the number of feature points meeting the conditions is less than 5, then only the feature points that actually meet the conditions are selected to determine the non-overlapping embedding region. In this preferred embodiment, after embedding the watermark according to the above steps, the peak signal-to-noise ratio (PSNR) of the watermarked image and the image to be embedded with the watermark is 40.23 dB, which basically meets the preset value of 40 dB. Generally, a higher PSNR indicates better visual quality of the image. When the PSNR reaches 40 dB or higher, the human eye cannot perceive the difference between the image before and after watermark embedding. Figure 2 and Figure 3As shown.

[0109] The following preferred embodiment illustrates in detail a method for enhancing local watermark extraction of feature points under perceptual constraints and guidance.

[0110] In this embodiment, one image is randomly selected as the watermarked image, such as... Figure 3 As shown, the image size is 512. 512.

[0111] SIFT feature point detection was performed on the watermarked image, resulting in 100 feature points.

[0112] The feature points are sorted from highest to lowest intensity, and the 10 non-overlapping regions located by the 10 feature points with the highest intensity are selected as the watermark extraction region. The ratio of the watermark extraction region size to the watermarked image size is 1 / 6, which is 85. 85.

[0113] Based on the 64-bit number of the watermark information, each watermark extraction region is divided into blocks, and the watermark extraction region is divided into 8×8 image blocks.

[0114] According to formula (1), the mean value of a certain block in the watermark extraction region is obtained in the spatial domain. The watermark is extracted sequentially from each image block using the watermark extraction formula (10), and finally the complete 64-bit watermark information is obtained.

[0115] The reason for selecting 10 feature points to locate the watermark extraction area is that, in practical applications, after various image processing steps, the intensity of the original feature points may change, making it difficult to determine all the original feature points and their corresponding regions using the original number. Increasing the number of selected feature points during watermark extraction can cover a wider range, reduce the risk of missing original feature points, and increase the probability of successfully extracting the watermark.

[0116] In this embodiment, 110 images of different styles, sizes, and scenes were randomly selected as images to be embedded with watermarks. The proposed method was used to embed the watermarks, resulting in 110 watermarked images. The stability of the feature points was evaluated by assessing the changes in feature point intensity before and after watermark embedding. The effectiveness of the PSNR-based quantization parameter model of this invention was evaluated by comparing the PSNR values ​​of the original image region and the watermarked region with preset PSNR values. The robustness of the watermark was evaluated by using the average extraction accuracy under different types of image attacks with different parameters; that is, the percentage of images from which the complete watermark could be correctly extracted out of all watermarked images.

[0117] For the 110 watermarked images obtained from the above watermark embedding method embodiment, the intensity of feature points in the watermarked images was detected and compared with the corresponding feature point intensities in the original images. The results of the change in feature point intensity before and after watermark embedding are as follows: Figure 4 As shown in the results, the present invention enhances the feature point fusion process during watermark embedding, avoiding the weakening effect of watermark embedding on feature point enhancement, effectively enhancing feature point strength and improving its stability.

[0118] For 110 watermarked images obtained from the above watermark embedding method embodiment, the relationship between the actual PSNR of the watermark embedding region and the preset PSNR was statistically analyzed. The experimental results are as follows: Figure 5 As shown, the PSNR of the embedded region is basically consistent with the preset value, which proves the effectiveness of the PSNR-based quantization parameter model in this invention.

[0119] For the 110 watermarked images obtained from the above watermark embedding method embodiment, various image attacks were applied, including: speckle noise with energy of 0.01 and 0.02, salt-and-pepper noise with energy of 0.01 and 0.02, median filtering of 3×3 and 5×5, mean filtering of 3×3 and 5×5, scaling attacks with parameters of 0.80 and 0.75, cropping with ratios of 0.1 and 0.2, and rotation attacks with parameter of 2. Then, a perceptual constraint and guided feature point enhancement local watermark extraction method embodiment was used to extract the watermark from the attacked watermarked images, and the results were compared with the original watermark. The average extraction accuracy of the 110 extracted watermarks is shown in Table 1.

[0120] Table 1

[0121]

[0122] From Table 1, Figure 2 , Figure 3 , Figure 5 The results show that, in the face of common image attacks, the embodiments of the present invention can basically extract 100% of the watermark. This proves that the present invention can not only effectively resist various image attacks and has strong robustness, but also ensure the invisibility of the watermark in each image.

[0123] Those skilled in the art will understand that the above assumptions are merely examples and should not be considered as an improper limitation of the present invention.

[0124] Although the steps in the above embodiments are described in the above order, those skilled in the art will understand that in order to achieve the effect of this embodiment, different steps do not need to be executed in such order. They can be executed simultaneously (in parallel) or in reverse order. These simple variations are all within the protection scope of this invention.

[0125] The technical solutions provided by the embodiments of the present invention have been described in detail above. Although specific examples have been used to illustrate the principles and implementation methods of the present invention, the descriptions of the above embodiments are only for the purpose of helping to understand the principles of the embodiments of the present invention; at the same time, those skilled in the art will make changes in specific implementation methods and application scope based on the embodiments of the present invention.

[0126] It should be noted that the flowcharts or block diagrams involved in this article are not limited to the form shown in this article, and they can also be divided and / or combined.

[0127] It should be noted that the markings and text in the accompanying drawings are only for the purpose of more clearly illustrating the present invention and should not be regarded as an improper limitation on the scope of protection of the present invention.

[0128] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in sequences other than those illustrated or described herein.

[0129] The term "comprising" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent in such process, method, article, or apparatus / device.

[0130] The various steps of this invention can be implemented using general-purpose computing devices. For example, they can be centralized on a single computing device, such as a personal computer, server computer, handheld or portable device, tablet device, or multiprocessor device, or they can be distributed across a network of multiple computing devices. They can be performed in a different order than shown or described herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Therefore, this invention is not limited to any particular hardware and software or combination thereof.

[0131] The method provided by this invention can be implemented using programmable logic devices or as computer program software or program modules (including routines, programs, objects, components, or data structures that perform specific tasks or implement specific abstract data types, etc.). For example, according to an embodiment of the invention, it can be a computer program product, running which causes a computer to perform the demonstrated method. The computer program product includes a computer-readable storage medium containing computer program logic or code portions for implementing the method. The computer-readable storage medium can be an internal medium installed in a computer or a removable medium that can be removed from the computer body (e.g., a storage device employing hot-swappable technology). The internal medium includes, but is not limited to, rewritable non-volatile memory, such as RAM, ROM, flash memory, and hard disks. The removable medium includes, but is not limited to, optical storage media (e.g., CD-ROMs and DVDs), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tapes or portable hard disks), media with built-in rewritable non-volatile memory (e.g., memory cards), and media with built-in ROM (e.g., ROM cartridges).

[0132] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A feature point enhancement local watermark embedding method under perceptual constraints and guidance, specifically including: Step S1: Perform SIFT feature point detection on the image to be embedded with the watermark to obtain several feature points; Step S2: Sort the feature points according to their intensity from largest to smallest, and select multiple non-overlapping regions located by the feature points with the highest intensity as the watermark embedding region; Step S3: Divide each watermark embedding region into blocks according to the length of the watermark information bits; Step S4: Embed a one-bit watermark in each block of each watermark embedding region using a watermark embedding strategy enhanced by fused feature points, including: Step S41: Determine the embedding parameters in the watermark embedding strategy enhanced by fusion feature points based on the pre-established PSNR-based quantization parameter model. Step S42: Classify the blocks according to whether they contain feature points. Blocks containing feature points are called central image blocks, and blocks without feature points are called non-central image blocks. Step S43: Under the guidance of JND, a one-bit watermark is embedded in each of the non-center image blocks using a quantization method; the quantization formula is shown in formula (1), the weight is modified as shown in formula (2), and the embedding formula is shown in formula (3): （1） （2） （3） in, i, j Indicates the index of the image patch; B i,j and B′ i,j These represent the non-central image blocks in the image to be watermarked and the watermarked image, respectively. i,j and u′ i,j They represent B respectively i,j and B′ i,j The pixel mean; l q The embedding parameter, also known as the quantization step size, is used to control the amount of pixel modification; w i,j This indicates the watermark position to be embedded; round() is a rounding function, mod represents the modulo operation; Jnd i,j It is image block B i,j The JND matrix for each pixel, M i,j This represents the pixel modification weight of the non-center image block; sum() is the summation function; Step S44: Guided by JND, embed a one-bit watermark in the central image block using an optimization function; the weight matrix is ​​shown in formula (4), and the optimization function is shown in formula (5): （4） （5） （6） （7） Where h is the length of the image patch, w is the width of the image patch, and B o The watermark image I to be embedded contains feature points p. o The central image block, Jnd o B o The JND matrix; M o The pixel modification weight of the central image block is represented; I′ is the watermarked image, B o ′ is the watermarked image with B o The corresponding image patch; min indicates taking the minimum value, ||M o (B o -B o ′)|| denotes the L2 norm; constraint (C.1) guarantees p o The intensity increases to the specified value; p o :maximum and p o :minimun respectively represent p o It is S centered on itself po The maximum and minimum intensity feature points within the neighborhood; D I (p o ) represents feature point p o The initial intensity value, D I’ (p o ) represents feature point p o The enhanced intensity value, ξ, represents the enhancement value of the feature point intensity; constraint (C.2) guarantees S po No new feature points were generated in the neighborhood; p max and p min S represents po The maximum and minimum feature strengths within the neighborhood, and the constraint (C.3) ensure the embedding of the watermark; Step S5: Replace the corresponding area in the image to be embedded with the watermark with the watermark-embedded area to obtain a watermarked image.

2. The feature point enhancement local watermark embedding method under perceptual constraints and guidance according to claim 1, characterized in that, Step S41 includes: Step S411: According to formula (1), assume that the modification amount of each pixel in the block is Δu i,j The mean square error (MSE) and PSNR of the entire watermark embedding region satisfy the following relationship: （8） Step S412: Use Δu i,j 2 Expected value l q 2 / 4 Estimate the MSE of the embedded region and substitute it into formula (8) to obtain the quantization parameter model based on PSNR as follows: （9） Step S413: Calculate the value of the embedding parameter using the given PSNR value according to formula (9).

3. A method for extracting local watermarks with enhanced feature points under perceptual constraints and guidance, the method being used to extract watermarks embedded using the local watermark embedding method with enhanced feature points under perceptual constraints and guidance as described in any one of claims 1-2, comprising the following steps: Step T1: Using the same method as in step S1, perform SIFT feature point detection on the watermarked image to obtain several feature points; Step T2: Using the same method as in step S2, sort the feature points according to their intensity from largest to smallest, and select the multiple non-overlapping regions located by the feature points with the highest intensity as the watermark extraction regions. Step T3: Using the same method as in step S3, divide each watermark extraction region into blocks according to the length of the watermark information bits; Step T4: Extract one watermark bit from each block of each watermark extraction region to obtain the complete watermark information. The extraction formula is as follows: （10） in, i, j Indicates the index of the image patch; This represents the calculated mean value of the image patch; This indicates the extracted watermark position; round() is a rounding function, and mod represents the modulo operation.

Images