A hidden image restoration method based on estimated value and possible value probability

Abstract

The present invention discloses a hidden image restoration method based on estimation and possible value probability. The method divides pixels into true value pixels and distorted pixels according to distortion values, calculates the convex hull of the true value pixels in the dynamic neighborhood for the distorted pixel points, divides the pixel points to be restored into internal points and external points of the convex hull according to the relative position relationship between the pixel points to be restored and the convex hull, and uses different methods to estimate different categories of distorted pixels, combines the estimated values ​​with the distribution of possible values ​​of the neighborhood pixels, uses parameters to control their proximity, and determines the final restoration value of the point to be restored in combination with the probability of the possible values. The hidden image restored by this method has a good visual effect and a high peak signal-to-noise ratio.

Description

Technical Field

[0001] The invention relates to the field of information security and image processing technology, and in particular to a hidden image recovery method based on estimation and possible probability. Background Art

[0002] The 21st century is a digital age. With the vigorous development of modern information and network technology, the Internet has become an integral part of people's daily lives and business activities. With the advent of the "Internet +" era, data on the Internet has become increasingly diversified and enriched. People work, study, and share global resources through the Internet. At the same time, due to the highly open characteristics of computer networks, their security protection mechanisms are relatively weak, and they will often inevitably be attacked, resulting in serious problems such as personal data leakage, property theft, and even state secrets leakage. If the existence of secret information can be concealed while transmitting information, the security of transmitted data will be greatly improved. Therefore, information hiding technology has been created.

[0003] Information hiding technology is also called data hiding. Its main work is to embed secret information into carriers such as images, texts or microparticle DNA to make the secret information invisible, and to transmit the secret information through the carrier in the public network. Information hiding technology can hide both communication behavior and communication content, so it has high research value.

[0004] Using digital images as carriers, we embed secret information into them to obtain secret images. Secret images are extremely vulnerable to various types of attacks such as noise and shearing, as well as interference from channel devices during network transmission. When the information hidden in them is also an image, it will also be damaged, and the degree of damage is even greater. The application of traditional image denoising algorithms, such as filtering algorithms, to recover image information extracted from secret images is not ideal. Therefore, in order to ensure the security of secret images, research on hidden image recovery has emerged, and it has shown considerable theoretical significance and development prospects. Summary of the invention

[0005] The purpose of the present invention is to provide a hidden image recovery method based on estimation and possible value probability in view of the shortcomings of the prior art. The hidden image recovered by this method has good visual effect and high peak signal-to-noise ratio.

[0006] The technical solution for achieving the purpose of the present invention is:

[0007] A hidden image recovery method based on estimated value and possible value probability comprises the following steps:

[0008] 1) Calculate the probability of possible values ​​of the points to be repaired: In information hiding where both the carrier and the hidden information are images, when the carrier is damaged, the hidden image will also be damaged. Suppose that m bits of pixels in the hidden image are damaged. These damaged bits may be 0 or 1, for a total of 2 m There are 2 situations m possible values, one of which is the value of the original pixel,

[0009] Assume that the pixels in the 3×3 neighborhood of the pixel point f to be repaired are f1, f2, f3, f4, f5, f6, f7, f8, calculate the possible values ​​of f1, f2, f3, f4, f5, f6, f7, f8 to obtain the possible values ​​of all neighborhood pixels, define the probability of the possible value of the pixel f to be repaired, and restore the hidden image based on the probability of the possible value of f. Because the visual impact of the image on the lower three bits of the pixel being damaged is small, only the possible values ​​of the lower three bits being damaged are close to each other. Therefore, when calculating the probability of the possible value of the pixel point f to be repaired, the situation where the lower three bits of the pixel point f to be repaired are damaged is ignored. Assume that n bits of the upper 5 bits of the pixel point f to be repaired are damaged, and the possible value is g. i , i = 1, 2, ..., t, where t = 2 n , g i The probability is p i , define g i The frequency of the neighboring pixel is the possible value of [g i -7,g i +7], denoted as h i , then:

[0010]

[0011] where i = 1, 2, ..., t, possible values ​​g i The greater the probability, the greater the i The greater the probability of being the original pixel value;

[0012] 2) Hidden image restoration based on estimation and possible value probability: Since the destruction of the lower three bits of the image pixels has little visual impact on the image, the pixels with only the lower three bits destroyed are defined as true value pixels. The true value pixels are not restored, and only the non-true value pixels are restored. The process is as follows:

[0013] 2-1) Calculate the estimated value of the point to be repaired: Detect the noise points in the image, mark the damaged bits of the hidden image pixels, and determine all the true value pixels: Calculate the convex hull of the true value pixels in the 3×3, 5×5 and 7×7 neighborhoods of the pixel point f to be repaired in turn. If a convex hull exists and the pixel point f to be repaired is inside the convex hull, the convex hull calculation ends and F=1 is set; otherwise, F=0 is set. If F=1, the true value pixels inside and on the boundary of the convex hull are used to estimate the pixel point f to be repaired. If F=0, the pixels in the 3×3 neighborhood of the pixel point f are used to estimate the pixel point f to be repaired.

[0014] For the pixel f to be repaired, the closer the distance between pixel q and the pixel f to be repaired is, the greater the impact on the pixel f to be repaired is. Assume that the Euclidean distance between the pixel f to be repaired and the pixel q is d(f,q), and the distance weight of the pixel q is defined as 1 / d(f,q).

[0015] The smaller the pixel q is damaged, the greater the reference significance for repairing the pixel f. Let the undamaged bits of pixel q be k1, k2, ..., k h ,make:

[0016]

[0017] Where 0≤k j ≤7, 1≤j≤h, L(q) is the weight of the neighborhood pixel q,

[0018] Assume that the true value pixels inside and on the boundary of the convex hull are q1,q2,...,q u , c(f) is the estimated value of the pixel f to be repaired, the pixel value of pixel q is v(q), if F = 1, use q1, q2, ..., q u Calculate the estimated value of the pixel point f to be repaired; otherwise, use f1,f2,f3,f4,f5,f6,f7,f8 to calculate the estimated value of the pixel point f to be repaired, then:

[0019]

[0020] 2-2) Hidden image restoration: Use the estimated value c(f) of the pixel to be restored to find the possible value with the highest probability that is closest to c(f) to restore the pixel to be restored. Use parameters to control the closeness between c(f) and the possible value of the pixel to be restored f, and determine the final restoration value of f based on the probability of the possible value, that is:

[0021] Suppose the possible value of the pixel f to be repaired is g i , for a given threshold T1, if g i satisfy:

[0022] |g i -c(f)|≤T1,

[0023] This means that the possible value g satisfies the condition i The difference with the estimated value is small, and the probability that these possible values ​​are true values ​​is large. Suppose the possible values ​​that meet the above conditions are The corresponding probability is make:

[0024] Where 1≤i≤r,

[0025] i * To meet the condition |g i -c(f)|≤T1, and the possible value subscript with the highest probability, due to is a possible value obtained by taking the damaged bit in the high 5 bits of the pixel to be repaired as 0 or 1. Let The possible values ​​of the destroyed bits in the lower 3 bits are 0 or 1, respectively, and are z1, z2, ..., z l ,make:

[0026] j * =argmin j (|z1-c(f)|,|z2-c(f)|,...,|z l -c(f)|), where 1≤j≤l,

[0027] j * for z1,z2,...,z l The subscript of the possible value that is the smallest difference from c(f) is taken is the final repair value of the pixel f to be repaired.

[0028] The hidden image restored by this method has good visual effect and high peak signal-to-noise ratio. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is a schematic diagram of a hidden image in an embodiment;

[0030] Figure 2 is a schematic diagram of a carrier image in an embodiment;

[0031] Figure 3 Schematic diagram of the pixel matrix of the neighborhood of the point to be repaired in the embodiment;

[0032] Figure 4 In the embodiment Figure 3 Schematic diagram of whether the pixel in the matrix is ​​credible;

[0033] Figure 5 Schematic diagram of the position relationship between the point to be repaired and the convex hull of the neighborhood credible points in the embodiment;

[0034] Figure 6Schematic diagram of a hidden image restored when the noise is 0.1 in the embodiment;

[0035] Figure 7 Schematic diagram of a hidden image restored when the noise is 0.3 in the embodiment;

[0036] Figure 8 It is a schematic diagram of a hidden image restored when the noise is 0.5 in the embodiment. DETAILED DESCRIPTION

[0037] The present invention is further described in detail below with reference to the accompanying drawings and examples, but the present invention is not limited thereto.

[0038] Example:

[0039] A hidden image recovery method based on estimated value and possible value probability comprises the following steps:

[0040] 1) Calculate the probability of possible values ​​of the points to be repaired: In information hiding where both the carrier and the hidden information are images, when the carrier is damaged, the hidden image will also be damaged. Suppose that m bits of pixels in the hidden image are damaged. These damaged bits may be 0 or 1, for a total of 2 m There are 2 situations m possible values, one of which is the value of the original pixel,

[0041] Assume that the pixels in the 3×3 neighborhood of the pixel point f to be repaired are f1, f2, f3, f4, f5, f6, f7, f8, calculate the possible values ​​of f1, f2, f3, f4, f5, f6, f7, f8 to obtain the possible values ​​of all neighborhood pixels, define the probability of the possible value of the pixel f to be repaired, and restore the hidden image based on the probability of the possible value of f. Because the visual impact of the image on the lower three bits of the pixel being damaged is small, only the possible values ​​of the lower three bits being damaged are close to each other. Therefore, when calculating the probability of the possible value of the pixel point f to be repaired, the situation where the lower three bits of the pixel point f to be repaired are damaged is ignored. Assume that n bits of the upper 5 bits of the pixel point f to be repaired are damaged, and the possible value is g. i , i = 1, 2, ..., t, where t = 2 n , g i The probability is p i , define g i The frequency of the neighboring pixel is the possible value of [g i -7,g i +7], denoted as h i , then:

[0042]

[0043] where i = 1, 2, ..., t, possible values ​​g iThe greater the probability, the greater the i The greater the probability of being the original pixel value,

[0044] like Figure 1 and Figure 2 As shown, in this example, the information bit of the pixel f to be repaired is 01101100, the credible bit is marked as 01101110, and the possible values ​​are {108, 100, 76, 68, 44, 36, 4, 12}. The possible values ​​of the neighboring pixels are counted. Taking the possible value 108 of the pixel f to be repaired as an example, the frequency of the possible values ​​in the range of [101, 115] and the possible value 108 are taken as the frequency of the possible value. From the possible values ​​of the neighboring pixels, we can get p(108)=8 / 29, p(100)=26 / 145, p(76)=34 / 145, p(68)=5 / 29, p(44)=1 / 29, p(36)=7 / 145, p(4)=2 / 145, p(12)=6 / 145;

[0045] 2) Hidden image restoration based on estimation and possible value probability: Since the destruction of the lower three bits of the image pixels has little visual impact on the image, the pixels with only the lower three bits destroyed are defined as true value pixels. The true value pixels are not restored, and only the non-true value pixels are restored. The process is as follows:

[0046] 2-1) Calculate the estimated value of the point to be repaired: Detect the noise points in the image, mark the damaged bits of the hidden image pixels, and determine all the true value pixels: Calculate the convex hull of the true value pixels in the 3×3, 5×5 and 7×7 neighborhoods of the pixel point f to be repaired in turn. If a convex hull exists and the pixel point f to be repaired is inside the convex hull, the convex hull calculation ends and F=1 is set; otherwise, F=0 is set. If F=1, the true value pixels inside and on the boundary of the convex hull are used to estimate the pixel point f to be repaired. If F=0, the pixels in the 3×3 neighborhood of the pixel point f are used to estimate the pixel point f to be repaired.

[0047] For the pixel f to be repaired, the closer the distance between pixel q and the pixel f to be repaired is, the greater the impact on the pixel f to be repaired is. Assume that the Euclidean distance between the pixel f to be repaired and the pixel q is d(f,q), and the distance weight of the pixel q is defined as 1 / d(f,q).

[0048] The smaller the pixel q is damaged, the greater the reference significance for repairing the pixel f. Let the undamaged bits of pixel q be k1, k2, ..., k h ,make:

[0049]

[0050] Where 0≤k j ≤7, 1≤j≤h, L(q) is the weight of the neighborhood pixel q,

[0051] Assume that the true value pixels inside and on the boundary of the convex hull are q1,q2,...,q u , c(f) is the estimated value of the pixel f to be repaired, the pixel value of pixel q is v(q), if F = 1, use q1, q2, ..., q u Calculate the estimated value of the pixel point f to be repaired; otherwise, use f1,f2,f3,f4,f5,f6,f7,f8 to calculate the estimated value of the pixel point f to be repaired, then:

[0052]

[0053] In this example, Figure 4 This is a schematic diagram of the matrix marking whether the neighboring pixels of the point to be repaired are credible. The pixel to be repaired is determined to be f. The distribution of the neighborhood pixel matrix of the hidden image to be repaired is as follows: Figure 3 As shown in the figure, set the threshold T1 = 4, first find the true value pixels in the 3 × 3 neighborhood of f, and get the true value pixels q7, q 16 ,q 18 , because the number of true value pixels is less than the threshold, the search range is expanded to the 4×4 neighborhood, and the true value pixels q4, q7, q 10 ,q 11 ,q 16 ,q 18 ,q 23 Since the number of true value pixels is greater than the threshold, stop searching and calculate the true value pixels q4, q7, q 10 ,q 11 ,q 16 ,q 18 ,q 23 The convex hull of the point f is related to the convex hull as follows: Figure 5 As shown by Figure 5 It can be seen that the point to be repaired is inside the convex hull, F = 1, so f is repaired based on the above pixel points:

[0054] Calculate the weight of the neighborhood true value pixel and estimate the point to be repaired. The weight is determined by the Euclidean distance between the point to be repaired f and the neighborhood true value pixel and the credibility of the true value pixel. Take the true value pixel q4 as an example. The 8-bit secret information of q4 is 01110100, and the credibility mark of the information bit is 00000101, where 1 indicates that the secret information is extracted from a noise point, and 0 indicates that the secret information is extracted from a non-noise point. The Euclidean distance between point q4 and point f is The credibility weight is calculated based on the credibility bit mark. The weight is defined as the product of the distance weight and the credibility weight, that is, Similarly, we can get L(q7)=0.7, L(q 10 )=0.45,L(q 11)=0.5, L(q 16 )=0.7,L(q 18 )=0.7,L(q 23 )=0.43. The estimated value of the point to be repaired f is 103;

[0055] 2-2) Hidden image restoration: Use the estimated value c(f) of the pixel to be restored to find the possible value with the highest probability that is closest to c(f) to restore the pixel to be restored. Use parameters to control the closeness between c(f) and the possible value of the pixel to be restored f, and determine the final restoration value of f based on the probability of the possible value, that is:

[0056] Suppose the possible value of the pixel f to be repaired is g i , for a given threshold T1, if g i satisfy:

[0057] |g i -c(f)|≤T1,

[0058] This means that the possible value g satisfies the condition i The difference with the estimated value is small, and the probability that these possible values ​​are true values ​​is large. Suppose the possible values ​​that meet the above conditions are The corresponding probability is make:

[0059] Where 1≤i≤r,

[0060] i * To meet the condition |g i -c(f)|≤T1, and the possible value subscript with the highest probability, due to is a possible value obtained by taking the damaged bit in the high 5 bits of the pixel to be repaired as 0 or 1. Let The possible values ​​of the destroyed bits in the lower 3 bits are 0 or 1, respectively, and are z1, z2, ..., z l ,make:

[0061] j * =argmin j (|z1-c(f)|,|z2-c(f)|,...,|z l -c(f)|), where 1≤j≤l,

[0062] j * for z1,z2,...,z l The subscript of the possible value that is the smallest difference from c(f) is taken is the final repair value of the pixel f to be repaired. In this example, the threshold T1 is set to 8, because |108-103|

Claims

1. A hidden image recovery method based on estimated value and possible value probability, characterized in that: The steps include: 1) Calculate the probability of possible values ​​of the points to be repaired: In information hiding where both the carrier and the hidden information are images, when the carrier is damaged, the hidden image will also be damaged. Suppose that m bits of pixels in the hidden image are damaged. These damaged bits may be 0 or 1, for a total of 2 m There are 2 situations m possible values, one of which is the value of the original pixel, Assume that the pixels in the 3×3 neighborhood of the pixel point f to be repaired are f1, f2, f3, f4, f5, f6, f7, f8. Calculate the possible values ​​of f1, f2, f3, f4, f5, f6, f7, f8 to get the possible values ​​of all the neighborhood pixels. Define the probability of the possible value of the pixel f to be repaired. According to the probability of the possible value of f, restore the hidden image. When calculating the probability of the possible value of the pixel point f to be repaired, ignore the situation where the lower three bits of the pixel point f to be repaired are damaged. Assume that n bits of the upper 5 bits of the pixel point f to be repaired are damaged, and the possible value is g i , i = 1, 2, ..., t, where t = 2 n , g i The probability is p i , define g i The frequency of the neighboring pixel is the possible value of [g i -7,g i +7], denoted as h i , then: where i = 1, 2, ..., t, possible values ​​g i The greater the probability, the greater the i The greater the probability of being the original pixel value; 2) Hidden image restoration based on estimation and possible value probability: The pixels of the image with only the lower three bits destroyed are defined as true value pixels. The true value pixels are not restored, and only the non-true value pixels are restored. The process is as follows: 2-1) Calculate the estimated value of the point to be repaired: Detect the noise points in the image, mark the damaged bits of the hidden image pixels, and determine all the true value pixels: Calculate the convex hull of the true value pixels in the 3×3, 5×5 and 7×7 neighborhoods of the pixel point f to be repaired in turn. If a convex hull exists and the pixel point f to be repaired is inside the convex hull, the convex hull calculation ends and F=1 is set; otherwise, F=0 is set. If F=1, the true value pixels inside and on the boundary of the convex hull are used to estimate the pixel point f to be repaired. If F=0, the pixels in the 3×3 neighborhood of the pixel point f are used to estimate the pixel point f to be repaired. For the pixel f to be repaired, the closer the distance between pixel q and the pixel f to be repaired is, the greater the impact on the pixel f to be repaired is. Assume that the Euclidean distance between the pixel f to be repaired and the pixel q is d(f,q), and the distance weight of the pixel q is defined as 1 / d(f,q). The smaller the pixel q is damaged, the greater the reference significance for repairing the pixel f. Let the undamaged bits of pixel q be k1, k2, ..., k h ,make: Where 0≤k j ≤7, 1≤j≤h, L(q) is the weight of the neighborhood pixel q, Assume that the true value pixels inside and on the boundary of the convex hull are q1,q2,...,q u , c(f) is the estimated value of the pixel f to be repaired, the pixel value of pixel q is v(q), if F = 1, use q1, q2, ..., q u Calculate the estimated value of the pixel point f to be repaired; otherwise, use f1,f2,f3,f4,f5,f6,f7,f8 to calculate the estimated value of the pixel point f to be repaired, then: 2-2) Hidden image restoration: Use the estimated value c(f) of the pixel to be restored to find the possible value with the highest probability that is closest to c(f) to restore the pixel to be restored. Use parameters to control the closeness between c(f) and the possible value of the pixel to be restored f, and determine the final restoration value of f based on the probability of the possible value, that is: Suppose the possible value of the pixel f to be repaired is g i , for a given threshold T1, if g i satisfy: |g i -c(f)|≤T1, This means that the possible value g satisfies the condition i The difference with the estimated value is small, and the probability that these possible values ​​are true values ​​is large. Suppose the possible values ​​that meet the above conditions are The corresponding probability is make: Where 1≤i≤r, i * To meet the condition |g i -c(f)|≤T1, and the possible value subscript with the highest probability, due to is a possible value obtained by taking the damaged bit in the high 5 bits of the pixel to be repaired as 0 or 1. Let The possible values ​​of the destroyed bits in the lower 3 bits are 0 or 1, respectively, and are z1, z2, ..., z l ,make: j * =argmin j (|z1-c(f)|,|z2-c(f)|,...,|z l -c(f)|), where 1≤j≤l, j * for z1,z2,...,z l The subscript of the possible value that is the smallest difference from c(f) is taken is the final repair value of the pixel f to be repaired.

Images