WO2021169436A1

WO2021169436A1 - Multi-layer digital veil design method for image content security and privacy protection

Info

Publication number: WO2021169436A1
Application number: PCT/CN2020/130722
Authority: WO
Inventors: 姚孝明
Original assignee: 海南大学
Priority date: 2020-02-26
Filing date: 2020-11-23
Publication date: 2021-09-02
Also published as: CN111310215A; ZA202201368B; CN111310215B

Abstract

A multi-layer digital veil design method for image content security and privacy protection, comprising single-layer digital veil generation, multi-layer digital veil generation, and digital veil installation and removal. In the present application, an existing pseudo-random sequence generator is used as a basic tool, the statistical characteristics of an image, and density parameters and strength parameters of a veil are selected according to protection requirements of the image content, and single-layer veil grid points of corresponding specifications and values thereof satisfying a given distribution are randomly generated, and a multi-layer veil is composed of single-layer veils of different specifications. After being fused with the image and then installed, the veil can effectively block most of details of a visual object in the image content while displaying hazy information of the overall content, and can destroy most of geometric features of the visual object, so as to effectively hide identity differences and local vision semantic differences of the visual object, and resist various image reconstruction attacks, such as high-and-low pass filtering and interpolation recovery, on damaged data or a noise signal, so that attackers cannot recover the image into a high-definition image and cannot implement content attacks such as "face changing".

Description

Multi-layer digital veil design method for image content security and privacy protection

Technical field

The invention belongs to the field of information technology, and specifically relates to a multi-layer digital veil design method for image content security and privacy protection.

Background technique

The widespread dissemination of multimedia content such as photos and images and the rapid development of artificial intelligence technology pose serious challenges to the privacy and security of image content. As a special form of unstructured data, image content includes visual objects (semantic objects with geometric structures), object attributes (including spatial location and size, color and texture, etc.), and object relationships (including structural relationships and visual objects Its characteristics, similarities and differences, and related relationships, etc.). Specifically, these contents are expressed in the image by group pixels and their mutual relations.

Existing image content and privacy protection technologies can be roughly divided into the following three types: (1) Encryption or selective encryption, that is, the image content is completely or partially encrypted or scrambled using traditional cryptographic methods, so that the image content in the corresponding area cannot be identified , So as to achieve the purpose of protecting image content and privacy; (2) Similar cartoon picture replacement method, that is, first segment the image visual object, and replace the sensitive area of the visual object (especially the face area) with the same cartoon picture, thereby hiding the visual object Some object attributes and relationships can achieve the purpose of protecting the privacy of a given visual object; (3) Digital watermarking method, which combines image reconstruction technology and adopts a variety of digital watermarking technologies to protect image content and private information, among which fragile watermarks can detect images Whether the content has been tampered with, the robust watermark can detect the copyright information of the image content, and the reversible watermark can be used to restore the difference between the original image and the reconstructed image.

Because the visual objects of the image content are related to each other, although the above-mentioned image content and privacy protection technology have progressed to varying degrees in recent years, basic problems still exist. Generally speaking, (1) Complete or selective encryption makes the image content of the corresponding area completely indistinguishable, and selective encryption has the problem of insufficient security. For example, Dahl et al. in 2017 gave the result of cracking the traditional "coding" (2) Although cartoon pictures are effective in hiding facial content, there is no guarantee that their identity cannot be inferred from other areas. For example, Ross et al. proposed soft feature recognition technology and gait based on biometric recognition in 2016 and 2018, respectively Feature recognition technology can obtain the identity information of the corresponding visual object based on related features such as gender, hairstyle, and gait; (3) Digital watermark technology can protect the copyright of the image content and whether it has been tampered with, and can also be combined with image reconstruction technology to protect image privacy information. The visual object is protected by "pretending to be absent", but this method puts forward extremely strict requirements on reversible watermarking technology and image compression technology, because the amount of information that reversible watermarking needs to hide at this time may be very high, especially for video images. .

In summary, the existing image content and privacy protection technologies have the following shortcomings: (1) Encrypting the entire or part of the image content, it is difficult to choose between the obscurity of the overall image content and the protection of the content and privacy, and lack of adjustment The flexibility of the two; (2) It is necessary to perform semantic segmentation on the visual objects of the image content. Under the condition of the existence of a large number of image samples, simple partial replacement is difficult to resist the attack of the correlation analysis between the content of the image area; (3) Implementation It is difficult and requires too strict requirements on the existing technology; (4) In addition, the disguise of the image content is not conducive to the promotion and dissemination of the image, and there are many unpredictable derivative social problems, such as a new form of fraud.

Summary of the invention

The purpose of the present invention is to provide a multi-layer digital veil design method for image content security and privacy protection. Through this method, single-layer and multi-layer digital veils can be made to block the specified part of the image, so that the attacker can only disguise the image. It can obtain the overall obscure information, but cannot obtain more local feature information, so as to effectively ensure the image content and privacy security.

In order to achieve the above objective, the technical solution of the present invention is to provide a multi-layer digital veil design method for image content security and privacy protection, including the following steps:

(1) Single-layer digital veil generation; the single-layer digital veil is divided into main veil and applied veil;

The main veil is generated by the following parameters: Given an M×N digital image matrix I, divide it into several m×n image sub-blocks; the veil density d is the percentage of the veil grid points in the overall image, or the veil rows and columns respectively account for the original The percentages of image rows and columns are determined by d1 and d2, that is, d=(d1,d2); the veil intensity parameter alpha≥1 is the adjustment parameter for the value of the veil grid point; the seed for generating the pseudo-random sequence z=(z1,z2,z3) ；

The veil is generated by the following parameters: the row and column sizes of a given digital image are M and N, respectively; the veil density d is the percentage of the veil grid points in the entire image, or is determined by the percentage of the veil rows and columns in the original image, d1 and d2. , That is, d=(d1,d2); the veil strength S is the maximum value of the grid points of the veil; the pseudo-random sequence seed key is z=(z1,z2,z3);

(2) Multi-layer digital veil generation; the multi-layer digital veil consists of a main veil and multiple veils, or only a plurality of veils;

(3) Installation and removal of the digital veil;

Installation of digital veil: From the original image to I, the updated image to I1, and the veil or multi-layer updated veil to V, the installation result Iv can be obtained as:

① Covered veil or multi-layered veil is Iv=I+V(mod 256);

②The main veil or the mixed veil of the main layer + overlay is Iv = I1 + V (mod 256);

Release of the digital veil: For the key owner, it is known that M, N, m, n, L, di, S (alpha), zi, P and Q, as well as the veil composition method and type parameters, especially if the reference strength Set by sub-block, the reference intensity setting value of each sub-block is also required. At this time, the digital veil can be released from the camouflage image Iv losslessly, and the original image I can be obtained.

Preferably, the process of generating the lay-up veil includes the following steps:

The row and column sizes of a given digital image are M and N respectively; the veil density d is the percentage of the veil grid points in the entire image, or is determined by the percentages of the veil row and column in the original image, d1 and d2 respectively, that is, d=(d1,d2 ); The veil strength S is the maximum value of the grid points of the veil; the pseudo-random sequence seed key is z=(z1,z2,z3); there are two ways to generate it, as follows:

(1) Method one

① Calculate the number of randomly selected grid rows and columns M1, N1

The number of veil rows M1=round(d1*M); the number of veil columns N1=round(d2*N)

Among them: round() is the rounding function

②Generate random grid position

a. Use the seed z1 to generate M pseudo-random numbers with a value ranging from 1 to M, and take the first M1 as the row coordinates of the grid;

b. Use the seed z2 to generate N pseudo-random numbers in the range of 1 to N, and take the first N1 as the column coordinates of the grid;

c. Expand the M1 row coordinates and N1 column coordinates according to the Cartesian product method to obtain the coordinate positions of all M1×N1 grid points;

③Generate the digital veil

Use seed z3 and veil strength S to generate a two-dimensional M1×N1 matrix with values ranging from 1 to S formed by M1×N1 pseudo-random integers, and assign the values in the matrix to the previously generated M1×N1 grid Click to complete the generation of a single-layer digital veil;

(2) Method two

①Calculate the randomly selected grid point position according to the given veil density

a. Use the seed z1 to generate an M×N random matrix R=[r _ij ] _M×N with a value range of (0,1) according to a certain random distribution;

b. Take the veil density d1 as the parameter of the threshold T, that is, let T = 1-d1, then the position of the grid point to be obtained is determined by _{the element of the matrix R*=[r* ij} ] _M×N with the value 1, where: If r _ij <T, then r* _ij =0; otherwise, r* _ij =1;

②Generate the digital veil

Using the seed z3 and the veil strength S to generate a uniformly distributed pseudo-random integer M×N matrix Vt ranging from 1 to S, then the single-layer digital veil V=Vt.*R*, that is, Vt and the grid point matrix R* Dot product (that is, the product of the corresponding elements of two matrices).

Preferably, the main veil generating process includes the following steps:

Given a M×N digital image matrix I, divide it into several m×n image sub-blocks; the veil density d is the percentage of the veil grid points in the entire image, or is determined by the percentages of veil rows and columns in the original image d1 and d2 respectively , That is, d=(d1,d2); the veil strength parameter alpha≥1 is the adjustment parameter for the value of the veil grid points; the seed z=(z1,z2,z3) for generating the pseudo-random sequence; the details are as follows:

(1) Pretreatment process

① Calculate the number of image blocks Br×Bc

Br=round(M/m); Bc=round(N/n); if mod(M,m)=a1≠0, mod(N,n)=a2≠0, you need to use 0 for the image matrix first Expand m-a1 rows and n-a2 columns below and on the right to get a (M+m-a1)×(N+n-a2) matrix; after the veil is generated, this part can be cropped to keep the image size consistent ；

Among them: round() is the rounding function.

②Calculate the mean P and mean square error Q of Br×Bc image sub-blocks

Extract each sub-block of the image according to the conventional method according to the progressive scanning method, calculate the mean value and mean square error of each sub-block, and save them to P and Q respectively;

③Calculate and update the image I1

According to the modulus 256 congruence algorithm, each sub-block of the original image I is subtracted from its mean value to obtain the updated sub-blocks and assembled according to their corresponding positions in the original image so that the element values are between 0 and 255, thereby obtaining the updated image I1;

④Setting and adjustment of sub-block veil strength

The veil strength S is equal to the adjustment parameter alpha multiplied by the reference strength (the result needs to be rounded to the nearest whole number), and its setting and adjustment have the following two processing options:

a. According to the protection requirements of each sub-block of the image, respectively specify its reference intensity, thereby determining S;

b. Take the mean square error of each sub-block as its reference intensity, thereby determining S.

(2) Generation of the main veil

① Select image sub-blocks A _ij in sequence, where: i = 1, 2,..., Br; j = 1, 2,..., Bc;

② Repeat this process for each sub-block to generate a single veil of corresponding density and strength;

According to the method of generating the veil, select method one or method two to generate the corresponding veil, where: in the second method, the grid random matrix is _{calculated by Vt=mod(Vt+round(mean2(A ij} )),256) ；

Note: round(), mean2() and mod() respectively refer to the corresponding rounding function, matrix mean function and modulo operation function in the matrix calculation software matlab7.1;

③The main veil is generated

Assemble each sub-veil according to the corresponding position of each sub-block in the original image, that is, the winner veil.

Preferably, the multi-faceted yarn generating process includes the following steps:

The multi-layer digital veil is composed of a main layer and multiple overlays, or only multiple overlays. The row and column sizes of a given digital image are M and N respectively, and the number of layers of multi-veil is L; the density of each veil is di=(di1,di2), the strength of each veil Si or the veil strength adjustment parameter is alphai, each layer selects The predetermined seed sequence is zi=(zi1, zi2, zi3), i=1,2,...,L; the multi-veil generation process is as follows:

(1) Initialization

Through simple assignment, an M×N veil matrix V with all 0 elements is generated;

(2) Loop L times to generate multi-veil

①According to the specification parameters di, Si or alphai, zi of each veil, use the above-mentioned single-veil generation process to obtain the single-veil matrix Vi;

②Update the veil matrix, namely V=Vi+V;

③ Use i to control steps ① and ② in this process for L times, where i=1, 2,...,L.

Preferably, the process of releasing the digital veil includes the following steps:

(1) To lay the veil: restore the original image I=Iv-V(mod 256);

(2) For the main layer or mixed veil:

①Restore and update the image I1=Iv-V(mod 256);

②Restore the original image I=I1+P (mod 256).

The so-called digital veil is to use the pseudo-random sequence generator as the basic tool to randomly select the grid points on the plane to form a veil grid on the image plane, and randomly allocate these grid points according to a certain common distribution (such as uniform distribution, normal distribution, etc.) The two-dimensional random integer matrix is formed by the integer value determined by the average value and the dispersion degree of the corresponding area of the image, and the magnitude is not limited to 256, which is fused with the original image, that is, the image is camouflaged after being installed in the original image.

The present invention does not need to pre-segment the visual object of the image content, but uses the existing pseudo-random sequence generator as the basic tool to select the statistical characteristics of the image, the density parameter and the intensity parameter of the veil according to the protection requirements of the image content, in a certain way Randomly generate corresponding single-veil grid points and their values that meet the given distribution. The multi-veil is composed of single-veil of different specifications, that is, a main layer and several layers. After the veil and the image are fused and installed, the overall display is displayed. The obscure content can effectively occlude most of the visual object details in the image content and destroy most of the visual object's geometric features, so as to effectively hide the identity differences and local visual semantic differences of visual objects, resist high and low pass filtering, interpolation restoration, etc. Image reconstruction attacks against lossy data or noise signals, selected plaintext attacks, only known ciphertext attacks, and noise removal attacks based on Generative Adversarial Networks (GAN), making it impossible for the attacker to recover high-definition images and implement "face swaps", etc. Content attacks provide solid technical support for multimedia content privacy protection and rights protection applications.

The present invention has made the following improvements and designs on the basis of the existing methods, namely 1) Based on visual objects as low-level features: although visual objects contain high-level semantic attributes such as shape, structure, color, and texture, in essence, their common features are: A random pattern formed by the population of pixels. Based on this fact, it is possible to determine the rationality of privacy protection that one random pattern is concealed by another random pattern; 2) Make full use of spatial modulus operations to adjust the distribution of randomness in space: In addition to reversibility, through modulo 256 Integer congruence operation can make the non-negative integers in the image approach the distribution of mean value 0, change the monotonic dependence of the veil intensity and the statistical characteristics of the regional dispersion, and make the partially homogeneous content approach inconsistent and partially heterogeneous content Convergence, thereby realizing statistically changing the geometric structure characteristics of visual objects; 3) Increasing veil parameter constraints to ensure the security against resampling attacks: Compressed sampling theory shows that precise pixels that are sparse enough can restore most of the other pixels in the image. In order to avoid the use of resampling methods such as band-limited filtering, interpolation, etc. to restore the original image, this method clearly requires that the number of unchanged pixels in the image after the veil installation should not exceed 0.008 as a whole according to the experimental results; 4) Design with full consideration of the statistical characteristics of image spatial distribution The veil ensures the relevance of the two and avoids the selection of plaintext attacks: the main veil structure design takes into account the complementary relationship between the low-frequency and high-frequency statistical characteristics of the image space to ensure the relevance of the veil and the image, so as to resist the selection of plaintext attacks; 5) Vision Adjustability: by randomly adjusting the veil grid and grid strength parameters, and if necessary, by adjusting the multi-layer composite parameters of the main and auxiliary veils, it has the visual adjustability of the image as a whole and the protected part.

Generally speaking, the digital veil needs to be merged with the original image to form the result image in the usual sense. This process is called the installation of the digital veil. After the veil is installed, the image is a disguised image; the process of non-destructively restoring the original image through a legal key and a small amount of secret information related to the image is the removal of the veil. Any attempt to recover the disguised image in whole or in part without legal authorization information is regarded as an illegal attack.

The multi-layer digital veil design method for image content security and privacy protection of the present invention has the following security:

① For an L-layer hybrid digital veil, its key space is not less than (M!N!/(M1!N1!)) ^L S ^M ^×N , and the statistical characteristics of the original image P, Q are used as the secret information sample The space is also close to 128 ^2×M×N . Therefore, when the veil density, veil strength, and number of layers are large enough, it has significant security and can resist ciphertext-only violent attacks.

② The sub-block mean matrix P and the mean square error matrix Q of the image I respectively express the low and high frequency characteristics of the original image. Since both the main layer and the hybrid veil contain P, the image must have relevant information after installation. Reduce the dimensionality of P and reduce the amount of effective information of P. In particular, the P information in the camouflage image changes randomly according to the distribution of the sub-block dispersion, so that the super-resolution restoration technology based on deep learning cannot be restored according to the "coding" condition. Can effectively resist super-resolution restoration attacks.

③The complexity of the digital veil is related to the complexity of the visual object of the image content to be protected. For visual objects with complex structure and texture, simple low-density veils can achieve better protection effects; for smooth and homogeneous visual objects, more complex high-density multi-veils are required. As long as the selection is appropriate, it can resist image restoration attacks such as interpolation reconstruction and low-pass filtering, making it difficult to reconstruct high-quality images without the corresponding key, and has a distinct active protection effect.

Compared with the prior art, the multi-layer digital veil design method for image content security and privacy protection of the present invention has the following beneficial effects:

1. High universality: The method of the present invention does not impose any restrictions on the visual content of images, allows various image visual content such as images, photos, videos, etc., and does not do anything about whether the content of the image itself is rich in texture or smooth areas, etc. Pre-setting, there is no need to segment the visual object in advance, so it has a higher universality.

2. Strong security: the key space is larger than (M! N!/(M1! N1!)) ^L S ^M×N , which is far more than the number of possible reconstructions of an image; on the other hand, as long as the veil density and veil If the intensity is set appropriately, the actual unaltered information of the camouflage image is often very small, not exceeding 2‰ of the overall image, making it difficult for illegal attackers to recover through image restoration technology or compressed sampling technology; therefore, it has superior security.

3. Strong adaptability: The digital veil and its number of layers and other parameters are relatively few, can be selected and adjusted according to the image content itself, and its effectiveness can be judged by the final image result, so it has strong content adaptability.

4. The program is simple and easy to implement: the digital veil uses a pseudo-random number generator as the basic tool, which is generated by random selection and random assignment of the image plane grid points. The algorithm is extremely simple, even if the multi-layer digital veil only needs L times The cycle can be completed, so it is extremely easy to implement.

Description of the drawings

Figure 1 is the standard test image lena, the image size is 512×512;

Figure 2 is a way to generate the main veil and the resulting image after installation;

Figure 3 shows the unchanged information after installation of the main veil in mode one and the result of interpolation recovery;

Figure 4 shows the results of unchanged information interpolation and low-pass and high-pass filter restoration after installation of the main veil in mode one;

Figure 5 is the second method to generate the main veil and its installation result image;

Figure 6 is the result image of the unchanged information of the main veil in mode two and its interpolation recovery;

Figure 7 shows the results of unchanged information interpolation and low-pass and high-pass filter restoration after the installation of the main veil in the second mode;

Fig. 8 is the result image of the generation and installation of the 5-layer mixed veil for the main layer method 1 and layer method 2 (the layer density increases from 0.3 to 0.2 to 0.9, and the intensity increases from 26 to 78);

Figure 9 The information and interpolation results of the 5 veils of the main layer mode 1 and lay-up mode 2 are unchanged after installation;

Figure 10 is an image of the result of low-pass and high-pass filtering of the camouflage image after the hybrid veil of Figure 8 is installed;

Figure 11 The resulting image of the generation and installation of 5 veils with both the main layer and the layup in mode two (the layup density increases from 0.3 to 0.2 to 0.9, and the intensity increases from 26 to 78);

Figure 12 is an image of unchanged information and its interpolation restoration result after the installation of the hybrid veil of Figure 11;

Figure 13 is an image of the result of low-pass and high-pass filtering of the camouflage image after the hybrid veil of Figure 11 is installed;

Fig. 14 is the low-pass and high-pass filtering results of the unchanged information interpolation image after the installation of the mixed veil of Fig. 11;

Fig. 15 is a histogram of images before and after installation of the hybrid veil of Fig. 11 (upper is the image histogram before installation, and the lower is the image histogram after installation).

Detailed ways

The specific implementation of the present invention will be described in further detail below in conjunction with examples. The following embodiments are only used to illustrate the present invention, but are not used to limit the scope of the present invention. The experimental methods without specific conditions in the following examples are usually in accordance with conventional conditions, or in accordance with the conditions recommended by the manufacturer.

1. The digital veil of the main layer: For the standard test image lena (see Figure 1), if the vertical and horizontal density is set to 0.992, the intensity condition parameter alpha is 5, and the first choice is to set the size of the lena image sub-block to 128×128, then The number of sub-blocks is 4×4=16, calculate the mean value and mean square deviation of these 16 sub-blocks to obtain P and Q. At this time, the veil intensity can be determined as round (Q*alpha), and then randomly determined by way of 508× 508 vertical and horizontal grid point coordinates, generate a 508×508 pseudo-random integer matrix with the veil intensity round (Q*alpha) as the amplitude, add the corresponding mean value P to the matrix and calculate the congruence value modulo 256 to generate a principal Veil, (In contrast, the second method uses the veil density as the threshold to obtain the grid point position R* that accounts for 99.2% of the 512×512 random matrix, and uses the veil intensity round (Q*alpha) as the amplitude to generate a 512×512 pseudo-random integer matrix, add the corresponding mean value P to the matrix and calculate the congruence value modulo 256 to determine the corresponding grid point value Vt, thereby obtaining the veil V) Install the veil to subtract each sub-block of lena respectively The updated image of the mean value is removed to obtain the installed image (see Figure 2 and Figure 5). Observing the veil (Fig. 2 and the left image of Fig. 5) can get the basic characteristics of the intensity distribution of the veil, with local homogeneity and inconsistency. The concept of digital veil is proposed to provide an effective tool for the differentiation of content area protection, that is, to control the processing strategy and corresponding parameters of image visual content protection based on the reverse matching of the original image and the digital veil in homogeneity and inconsistency, so that the image visual Content protection is more scientific and objective. Figures 2 and 5 are the resulting images after the single veil is installed on the protected image. It can be seen that most of the attributes and interrelationships of the visual objects are effectively hidden, leaving only the dim image of the overall image for fuzzy classification For management purposes or for interested users to purchase previews.

2. Multi-layer mixed digital veil: Fig. 8 and Fig. 11 respectively show a 5-layer mixed veil whose main layer is mode one, mode two, and main layer coatings are mode two, and the results after installation. Among them: the main veil density is 0.982, alpha=4.5, the lay density increases from 0.3 to 0.9, and the intensity increases from 26 to 78. It can be seen that the modification effect of the coating on the main layer and the partial hiding effect are also more delicate. In fact, Figure 15 shows the image histogram difference before and after the hybrid veil of Figure 11 is installed. The pixel distribution of the latter tends to be generally uniform, but some minor inconsistencies remain.

3. Ability to resist linear interpolation attacks: Fig. 3, Fig. 6, Fig. 9 and Fig. 12 respectively show the unchanged information of the camouflage images of the above specifications and their interpolation results. Under the condition of empirical knowledge, the human eye cannot see the correlation between the results generated by these interpolations and the original image.

4. Ability to resist low-pass filtering attacks: Figure 4, Figure 7, Figure 10, Figure 13, and Figure 14 respectively show the unaltered information interpolation image of various camouflage images and low-pass and high-pass filtering of camouflage images The result of the attack. It can be seen that the filtering attack does not provide additional image information.

Example

The following takes a simple image privacy protection application as an example to further illustrate the specific implementation results. In this application, on the one hand, it is required to disclose the basic content semantics of the image to facilitate classification, management, transmission, etc., on the other hand, it is required to hide the attribute characteristics of the image visual object such as gender and facial features.

The specific experimental operating environment is the operating system windows7, and the simulation experiment software is matlab7.1.

First select the standard test image lena, whose image size is 512×512, as shown in Figure 1. According to the requirements for the protection of visual objects, it is planned to choose 5 layers of mixed veil, of which the main layer and the lay-up layer are both realized in the second method. In order to ensure that part of the image information can be displayed dimly, set the main veil density to 0.982, and the veil density to 0.3, 0.5, 0.7 and 0.9 respectively, which are all lower than the minimum unchanged information requirement of 0.992, so that the image information is sufficiently retained, but After the final veil is installed, the overall covering density reaches 0.9943, which meets the corresponding safety requirements. It is also possible to balance the overall image hiding situation by comprehensively choosing between the main veil strength adjustment parameter alpha and the layer strength strength. For example, let alpha=4.5, and the strength is uniformly proportional to the layer density, that is, strength=density*87. Get the desired mixed veil and its image after installation, as shown in Figure 11. It is not difficult to see that, with the exception of “hats” and “faces” that are faintly identifiable as a whole, most of the features of hats and faces, including gender characteristics, have been successfully concealed, which obviously meets the needs of practical applications.

The PSNR value of the image and the original image restored by the above-mentioned test system against the interpolation reconstruction attack of the single-layer digital veil and the multi-layer digital veil and the low-pass filtering and the original image are shown in the following table.

The above-disclosed are only preferred embodiments of the present invention. Of course, the scope of rights of the present invention cannot be limited by this. Therefore, equivalent changes made according to the claims of the present invention still fall within the scope of the present invention.

Claims

A multi-layer digital veil design method for image content security and privacy protection, which is characterized by including:

(1) Single-layer digital veil generation; the single-layer digital veil is divided into main veil and applied veil;

The main veil is generated by the following parameters: Given an M×N digital image matrix I, divide it into several m×n image sub-blocks; the veil density d is the percentage of the veil grid points in the overall image, or the veil rows and columns respectively account for the original The percentages of image rows and columns are determined by d1 and d2, that is, d=(d1,d2); the veil intensity parameter alpha≥1 is the adjustment parameter for the value of the veil grid point; the seed for generating the pseudo-random sequence z=(z1,z2,z3) ；

The veil is generated by the following parameters: the row and column sizes of a given digital image are M and N, respectively; the veil density d is the percentage of the veil grid points in the entire image, or is determined by the percentage of the veil rows and columns in the original image, d1 and d2. , That is, d=(d1,d2); the veil strength S is the maximum value of the grid points of the veil; the pseudo-random sequence seed key is z=(z1,z2,z3);

(2) Multi-layer digital veil generation; the multi-layer digital veil consists of a main veil and multiple veils, or only a plurality of veils;

(3) Installation and removal of the digital veil;

Installation of digital veil: From the original image to I, the updated image to I1, and the veil or multi-layer updated veil to V, the installation result can be:

① Covered veil or multi-layered veil is Iv=I+V(mod 256);

②The main veil or the mixed veil of the main layer + overlay is Iv = I1 + V (mod 256);

The release of the digital veil: For the key owner, it is known that M, N, m, n, L, d, S (alpha), zi (i = 1, 2, 3), P and Q, and the way the veil is constructed are the same The type parameter, especially, if the reference intensity is set by sub-block, the reference intensity setting value of each sub-block is also required. At this time, the digital veil can be released from the camouflage image Iv without loss, and the original image I can be obtained.
The multi-layer digital veil design method for image content security and privacy protection according to claim 1, wherein the process of generating the veil includes the following steps:

① Calculate the number of randomly selected grid rows and columns M1, N1

The number of veil rows M1=round(d1*M); the number of veil columns N1=round(d2*N);

②Generate random grid position

a. Use the seed z1 to generate M pseudo-random numbers with a value ranging from 1 to M, and take the first M1 as the row coordinates of the grid;

b. Use the seed z2 to generate N pseudo-random numbers in the range of 1 to N, and take the first N1 as the column coordinates of the grid;

c. Expand the M1 row coordinates and N1 column coordinates according to the Cartesian product method to obtain the coordinate positions of all M1×N1 grid points;

③Generate the digital veil

Use seed z3 and veil strength S to generate a two-dimensional M1×N1 matrix with values ranging from 1 to S formed by M1×N1 pseudo-random integers, and assign the values in the matrix to the previously generated M1×N1 grid Click to complete the generation of a single-layer digital veil.
The multi-layer digital veil design method for image content security and privacy protection according to claim 1, wherein the process of generating the veil includes the following steps:

①Calculate the randomly selected grid point position according to the given veil density

a. Use the seed z1 to generate an M×N random matrix R=[r ij ] M×N with a value range of (0,1) according to a certain random distribution;

b. Take the veil density d1 as the parameter of the threshold T, that is, let T = 1-d1, then the position of the grid point to be obtained is determined by the element of the matrix R*=[r* ij ] M×N with the value 1, where: If r ij <T, then r* ij =0; otherwise, r* ij =1;

②Generate the digital veil

Using the seed z3 and the veil strength S to generate a uniformly distributed pseudo-random integer M×N matrix Vt ranging from 1 to S, then the single-layer digital veil V=Vt.*R*, that is, Vt and the grid point matrix R* Dot product is the product of the corresponding elements of two matrices.
The multi-layer digital veil design method for image content security and privacy protection according to claim 1, wherein the main veil generation process includes the following steps:

(1) Pretreatment process

① Calculate the number of image blocks Br×Bc

Br=round(M/m); Bc=round(N/n); if mod(M,m)=a1≠0, mod(N,n)=a2≠0, you need to use 0 for the image matrix first Expand m-a1 rows and n-a2 columns below and on the right to get a (M+m-a1)×(N+n-a2) matrix; after the veil is generated, this part can be cropped to keep the image size consistent ；

②Calculate the mean P and mean square error Q of Br×Bc image sub-blocks

Extract each sub-block of the image according to the conventional method according to the progressive scanning method, calculate the mean value and mean square error of each sub-block, and save them to P and Q respectively;

③Calculate and update the image I1

According to the modulus 256 congruence algorithm, each sub-block of the original image I is subtracted from its mean value to obtain the updated sub-blocks and assembled according to their corresponding positions in the original image so that the element values are between 0 and 255, thereby obtaining the updated image I1;

④Setting and adjustment of sub-block veil strength

The veil strength S is equal to the adjustment parameter alpha multiplied by the reference strength, and its setting and adjustment have the following two processing options:

a. According to the protection requirements of each sub-block of the image, respectively specify its reference intensity, thereby determining S;

b. Take the mean square error of each sub-block as its reference intensity, thereby determining S;

(2) Generation of the main veil

① Select image sub-blocks A ij in sequence, where: i = 1, 2,..., Br; j = 1, 2,..., Bc;

② Repeat this process for each sub-block to generate a single veil of corresponding density and strength;

Follow the steps of applying the veil to generate the corresponding veil;

③The main veil is generated

Assemble each sub-veil according to the corresponding position of each sub-block in the original image, that is, the winner veil.
The multi-layer digital veil design method for image content security and privacy protection according to claim 1, wherein the multi-layer digital veil generation process includes the following steps:

The row and column sizes of a given digital image are M and N respectively, and the number of layers of multi-veil is L; the density of each veil is di=(di1,di2), the strength of each veil Si or the veil strength adjustment parameter is alphai, each layer selects The predetermined seed sequence is zi=(zi1, zi2, zi3), i=1,2,...,L; the multi-veil generation process is as follows:

(1) Initialization

Through simple assignment, an M×N veil matrix V with all 0 elements is generated;

(2) Loop L times to generate multi-veil

①According to the specification parameters di, Si or alphai, zi of each veil, use the above-mentioned single-veil generation process to obtain the single-veil matrix Vi;

②Update the veil matrix, namely V=Vi+V;

③ Use i to control steps ① and ② in this process for L times, where i=1, 2,...,L.
The multi-layer digital veil design method for image content security and privacy protection according to claim 1, wherein the process of removing the digital veil includes the following steps:

(1) To lay the veil: restore the original image I=Iv-V(mod 256);

(2) For the main layer or mixed veil:

①Restore and update the image I1=Iv-V(mod 256);

②Restore the original image I=I1+P (mod 256).