CN108648133A - A kind of rotation of combined block and mosaic without embedded camouflage method - Google Patents

Abstract

What the present invention provided a kind of rotation of combined block and mosaic pretends reconstructing method without embedded, it is first the circular image for being r with pseudorandom knuckle radius by the greyscale image transitions that multiple resolution ratio are (2r 1) × (2r 1), stowed position of the bit bit string of close figure division on blindage is determined by generating random integers coordinate sequence；Secondly the bit bit string hidden and non-concealed position is then associated with respectively with key and position by placement or divides close figure with the immediate circular image of blindage placement location is carried out coded representation and is formed to cover, and add and the circular image of placement is authenticated with key and position associated random corner；The deviation that placement process generates finally is passed into around blindage unprocessed pixel to generate mosaic image and extract Mi Tu from mosaic image according to key and be authenticated to the close figure extracted.Institute's extracting method can carry out close figure the mosaic cover of high quality and can resist channel attack and be strictly dependent on key have higher safety.

Description

An embedding-free camouflage method combining block rotation and mosaic

technical field

The invention belongs to the cross field of image information security and digital image signal processing, relates to a camouflage method, in particular to a dense image camouflage and restoration method combined with block rotation and mosaic.

Background technique

At present, with the continuous deepening of deep learning, the continued development of artificial intelligence and the emergence of quantum computers, the traditional multimedia information security situation with image and audio as the main transmission medium has become more severe. At the same time, on the one hand, the continuous development of compression technology also makes the redundant space available for traditional modified embedding-based information hiding smaller and smaller; on the other hand, the steganographic classifier dimension based on machine learning The number of data continues to increase, and even a 34,761-dimensional spatial domain-rich feature model appears, which makes it less and less likely that the hidden information of traditional information hiding will not be discovered, and all these also make the development of traditional information hiding technology based on modified embedding fall into a trap bottleneck.

How to effectively research the next generation of information hiding technology, experts from Beijing and Shanghai held the National Symposium on Information Hiding and Multimedia Security Experts held in May 2014 for the first time to put forward "carrier-free information hiding". At the 12th National Conference on Information Hiding held in Wuhan in March 2015, researcher Guo Yunbiao, director of the Beijing Electronics Technology Application Research Institute, was invited to report at the conference: My opinion on information hiding, including carrierless information hiding technology in future information Hidden frontier. The 13th National Information Hiding Conference held in Hefei in October 2016 officially positioned carrierless information hiding as the second generation of information hiding technology, and the two conference theme reports are directly related to carrierless information hiding. There are two typical methods based on traditional blanking information hiding technology, which are search-based blanking information hiding method and generative blanking information hiding method based on texture synthesis.

Traditional search-based carrierless information hiding methods mainly transmit secret information by retrieving carrier text or images containing specified secret vectors in the database. Due to the limited ability of natural image text to express irrelevant secret information, the embedding of such methods Very low density. For example, Zhou Z L, 2015. (Zhou Z L, Sun H Y, Harit RH, Chen X Y, Sun X M. Coverless image steganography without embedding[C]//International Conference on Cloud Computing and Security. Springer International Publishing, 2015:123-132 .) Based on the comparison of the block mean value, the image is divided into 9 blocks, and the image is further mapped to the corresponding 8-bit hash value by means of the block mean value comparison, and finally the image in the database whose hash value is equal to the secret bit segment is retrieved as With a dense carrier, this method can only hide 8 bits per image. Zhou Z,2017(Zhou Z,Wu Q M J,Yang C N,et al.Coverless image steganography using histograms of oriented gradients-based hashing algorithm[J].Journal of Internet Technology,2017,18(5):1177-1184) in Zhou Z L , 2015. Based on the introduction of user identification to determine the position of the image block where the secret information is embedded, and further map the pixel gradient magnitude and gradient direction of the image block to 20 bits, so the embedding capacity is only 20 bits per image. In order to further increase the embedding capacity of a single image and reduce the size of the database, Wu Jianbin, 2018 (Wu Jianbin, Jia Yanke, Liu Yiwen. Carrier-free information hiding based on image coding and splicing [C]//The 14th National Information Hiding and Multimedia Information Security Academic Conference (CIHW2018, Guangzhou, 2018: 45-52) encodes the information entropy of the image into a 7-bit hash value through the hash function, and stitches the 4 images together by permutation and combination, using the hash value expressed by the 4 images Hidden values and 16 splicing sequences of 4 images hide 32 bits of information. Due to the extremely low embedding capacity of the search-based carrierless information hiding method, a large amount of carrier text or images need to be transmitted in the channel to fully express the secret information, which easily arouses the suspicion of the attacker, and the transmitted secret information is easily damaged.

The main idea of the traditional generative carrierless information hiding method based on texture synthesis is to express secret information through texture generation. For example, Wu K C, 2015 (Wu K C, Wang C M. Steganography using reversible texture synthesis[J]. IEEE Transactions on Image Processing. 2015, 24(1): 130-139) divides simple texture texture images into non-overlapping equal-sized Image small blocks, using MSE to divide image small blocks into different levels. When embedding, the secret information is corresponding to the MSE level of the small image block, and the appropriate candidate small block is selected to form the embedding mask. Another example, Qian Z X, 2017 (Qian Z X, Zhou H, Zhang W M, et al. Robuststeganography using texture synthesis. Advances in Intelligent Information Hiding and Multimedia Signal Processing. Smart Innovation, Systems and Technologies, 2017, 63:25-33) according to small The complexity of the core area of the block classifies the small blocks, and each classification represents a kind of secret information, and the selected small blocks are randomly placed in the cover image, and the secret information is concealed by splicing them with other small blocks. However, such methods can only generate texture images with simple textures, and it is difficult to generate complex and meaningful images.

Traditional mosaic-based information hiding methods can produce meaningful images, but they need to modify the carrier to embed secret information. For example, Lai I J,2011(Lai I J,Tsai W H.Secret-fragment-visible mosaic image–a new computer art and its application to informationhiding[J].IEEE Transactions on Information Forensics&Security,2011,6(3):936- 945) to find the cover image that is as large and most similar to the secret map, use the secret map to divide small blocks to splice the public image, and use the reversible LSB embedding method to embed the mapping parameters, so as to transfer the secret map with the public image after splicing, but this The method is limited to disguising the dense image as a public image with a visual quality close to it, resulting in a narrow range of application. In order to improve the visual quality of public images after masquerading and the restoration quality of secret images, and to avoid free selection of public images, Zhai S Y, 2015 (Zhai SY, Li F, Chang C C, et al. A meaningful scheme for sharing secret images using mosaic images[ J].International Journal of Network Security,2015,17(5):643-649) through the method of replacing the small block of the public image with the small block of the dense image, the dense image is hidden in 4 randomly selected bunker images, and The LSB method is used to embed the recovery parameters of the dense graph. Zhang Meng, 2016 (Zhang Meng, Zhai Shengyun, Su Dongqi. Improved secret image sharing algorithm based on mosaic technology [J]. Computer Application Research, 2016, 33(11): 3480-3484) Combined with the difference expansion reversible information hiding method, based on The method of mosaic mosaic provides a camouflage strategy from the dense image to the public image, so that the small blocks of the dense image can be completely restored after parameter extraction. Lee Y L,2014(Lee Y L,Tsai W H.A new secure imagetransmission technique via secret-fragment-visible mosaic images by nearly reversible color transformations[J].IEEE Transactions on Circuits&Systems for Video Technology,2014,24(4):695-703) The secret image patch and the cover image patch are sorted according to the mean and standard deviation respectively, and then a one-to-one mapping relationship is established according to the sorting order, and an angle transformation is introduced to improve the visual matching quality of the secret image patch to the public image patch. Hou D D, 2016 (Hou D, Zhang W, Yu N. Image camouflageby reversible image transformation [J]. Journal of Visual Communication & Image Representation, 2016, 40: 225-236) introduced K-means clustering pair based on Lee Y L, 2014 The dense image patches and cover patches are classified and matched, and the embedding parameters are optimized. Liu Xiaokai, 2018 (Liu Xiaokai, Yao Heng, Qin Chuan. An improved reversible image camouflage based on image block classification threshold optimization [C]//The 14th National Conference on Information Hiding and Multimedia Information Security (CIHW2018), Guangzhou, 2018: 37-44) made improvements to Hou D D, 2016, introduced the classification threshold optimization algorithm to make the mean square error between the matched dense map small block and the mask small block smaller, and further added horizontal flip in the angle transformation, so that The resulting public images have better visual quality.

In addition to the mosaic puzzle that divides the secret image and the public image into small pieces for splicing, there is also an image mosaic puzzle method that stitches small images into large images. The local details of this puzzle method are small images, and all the small images can be organized together to form a complete other image with any expressive force. Based on the image mosaic puzzle method, Lin W L, 2004 (Lin W L, Tsai W H. Data hiding in image mosaics by visible boundary regions and its copyright protection application against print-and-scan attacks[J].2004) divided the mask image into blocks, Then each block is corresponding to a rectangular image, and the variance of the left, right, upper and lower sides of the rectangular image is adjusted by adding noise to determine the 2-bit secret information represented, and the secret information is extracted according to the minimum side variance greater than the threshold.

The mosaic-based information hiding methods given above all involve the modification of the cover image, which will leave traces of modification in the carrier. At the same time, the parameter embedding method given is less robust, and when attacked, It is easy to cause the loss of embedded parameters, so that the authenticity and reliability of the reconstructed dense map cannot be accurately identified.

Contents of the invention

The purpose of the present invention is to overcome the defects of the prior art, and propose a non-embedded camouflage reconstruction method combining block rotation and mosaic, using image mosaic, based on the method of direct generation of encrypted carrier, using arbitrarily selected images to generate open and meaningful In the generation process, the randomly selected image is converted into a circular image, and the circular image with random corners is used to express different secret information at different positions. The proposed method is strictly dependent on the user key. Only the provided The correct key is required to obtain the secret information and verify the accuracy of the extracted secret information, which cannot be obtained otherwise. Since the circular image only involves random corners and does not involve any parameter modification embedding, it has strong robustness.

To achieve the above object, the present invention adopts the following technical solutions:

A non-embedding camouflage method combining block rotation and mosaic, characterized in that it comprises the following steps:

Step 1: Input a P ₀ -level grayscale mask image with a resolution of m ₀ ×n ₀ P ₁ -level grayscale secret image with resolution m ₁ ×n ₁ And l P ₂ -level grayscale images with a resolution of (2r-1)×(2r-1) in pairs k=0,1,...,l-1, among them, P ₀ , P ₁ are integers greater than or equal to 0, r is a positive integer greater than 0, l is an integer power of 2 and satisfies

Step 2: Generate a random integer ind _k ∈ [0,n-1] from the key k ₀ , k=0,1,...,l-1, and n is an integer greater than 0;

Step 3: Map ind _k to a random rotation angle α _ind between [0,2π), then convert all H _k into circular images with random rotation angle α _ind and radius r

Step 4: Scan S as a sequence of 2-valued bits Generate l ₀ random integer coordinate sequences in the range of m ₀ ×n ₀ from the key k ₁ in, is the ith random integer coordinate, namely

Step 5: Initialize the P ₀ -level grayscale mask image with a resolution of m ₂ ×n ₂ And satisfy m ₂ =m ₀ (2r-1), n ₂ =n ₀ (2r-1), generate m ₀ n ₀ random integer matrices in the range of [0,l-1] from key k ₂ And r _{i, j} ∈ [0, l-1];

Step 6: For Intercept the 2-value bit B _cur from B, convert it to an integer X _Cur , and further map X _Cur to an integer in the range [0,l-1] Then select the first from H′ _k , k=0,1,…,l-1 images and calculate The random rotation angle α _ind will be Rotate counterclockwise to Place it on the matrix block selected on M with (x·(2r-1), y·(2r-1)) as the starting point of the upper left corner and a size of (2r-1)×(2r-1);

Step 7: Calculate The difference Δx _{,y between the pixel t x,y and} the grayscale mask image T, and then pass Δx, _y to the unprocessed pixels around t _x,y

Step 8: Repeat steps 6 to 7 until all coordinate positions in T ₀ are processed;

Step 9: For From H′ _k , k=0,1,…,l-1, find the closest small block to pixel t _{x, y} on T and calculate The random rotation angle α _ind will be Rotate counterclockwise to Place it on the matrix block selected on M with (x·(2r-1), y·(2r-1)) as the starting point of the upper left corner and a size of (2r-1)×(2r-1);

Step 10: Calculate The difference between t _x,y and Δ _x,y , then pass Δ _x,y to the unprocessed pixels around t _x,y

Step 11: Repeat steps 9 to 10 until all coordinate positions that do not belong to T ₀ are processed;

Step 12: Use the final M output as an embedding mosaic image.

Preferably, in the third step, ind _k is mapped to a random rotation angle α _ind between [0,2π), and then all H _k are converted into a circular image with a random rotation angle α _ind and a radius of r The specific method is:

Transform all H _k into circular images with random rotation angle α _ind and radius r according to formula (1) Among them, α _ind = ind _k · 2π/n;

H′ _k =Rot(H _k ,r,α _ind ) (1);

In formula (1), Rot() is a counterclockwise rotation function, H _k corresponds to the input original image, r is the radius of the output ring image, and α _ind is the counterclockwise rotation angle.

Preferably, the specific implementation steps of the Rot() function in formula (1) are shown in steps 3.1 to 3.4:

Step 3.1: Initialize H′ _k according to formula (2):

Step 3.2: Pixels within H _k radius r Convert to polar coordinates according to formula (3) Will Rotate the angle α _ind counterclockwise and transform it into Cartesian coordinates (i ₁ , j ₁ ) according to formula (4), then Assign to elements in H′ _k Among them, when i=r-1, j=r-1, then directly as

In formula (4), [] is a rounding operator;

Step 3.3: After each Find within the centered 5x5 field The pixels of which are not -1 are recorded as P ₀ , P ₁ ,…,P _count-1 , and the corresponding and The secondary distances of are recorded as d ₀ ,d ₁ ,…,d _count-1 in turn, and then use formula (5) to Perform interpolation fitting;

Step 3.4: Output H′ _k .

Preferably, in step 6, the specific method of intercepting the 2-value bit B _cur from B is:

Intercept log ₂ l binary bits B _cur from B according to formula (6), if the total number of intercepts is less than log ₂ l, then intercept all remaining bits;

B _cur = Trim(B, Cur·log ₂ l, log ₂ l) (6);

In formula (6), Trim() is a binary sequence truncation function, where the first input parameter of the Trim() function is the input binary bit sequence, the second parameter is the starting index position of the interception, and the third parameter corresponds to is the truncation length, Cur is the binary bit index of the current truncation and Cur≥0;

In step 6, map X _Cur to an integer in the range [0,l-1] The specific method is formula (7):

In steps 6 and 9, the random rotation angle α _ind is calculated, and the Rotate counterclockwise to The specific method is:

Calculate the integer ind in the range of [0, n-1] according to formula (8), and then use formula (1) to Rotate α _ind counterclockwise to get where α _ind =ind·2π/n:

ind = (X _Cur + x y + y r _{x, y} + x r _{x, y} ) mod n (8);

H′ _k =Rot(H _k ,r,α _ind ) (1);

Preferably, in step 9, for From H′ _k , k=0,1,…,l-1, find the closest small block to pixel t _{x, y} on T The specific method is:

Find the small block closest to the average value of pixel t _{x, y} on T from H′ _k ,k=0,1,…,l-1 according to formula (9)

In formula (9), mean() is a circular image block mean value calculation function, which calculates the mean value of pixels within a circular image range whose radius is r with the center of the input image as the center;

In step 10, calculate The specific method of the difference Δ _{x, y between t x, y} is formula (10):

In step 10, pass Δ _x,y to the unprocessed pixels around t _x,y The specific method is formula (11):

A non-embedded restoration method combining block rotation and mosaic corresponding to claim 1, comprising the following steps:

Step 1: Input a P ₀ -level grayscale mosaic image with a resolution of m ₂ ×n ₂ l P ₂ -level grayscale images with a resolution of (2r-1)×(2r-1) in pairs Secret image resolution m ₁ ×n ₁ , gray scale P ₁ of secret image, initialize binary bit sequence B=Φ and B′=Φ, input key k ₀ , k ₁ , k ₂ and set threshold parameter T ,T>0;

Step 2: Generate a random integer ind _k ∈ [0,n-1] from the key k ₀ , k=0,1,...,l-1 and n is an integer greater than 0, and map ind _k to [0, 2π) with a random rotation angle α _ind , and then convert all H _k into a circular image with a random rotation angle α _ind and radius r

Step 3: Correct all H′ _k rotations to H″ _k and satisfy the H″ _k centroid in the direction agreed by the H″ _k circle center;

Step 4: Initialize the P ₁ -level grayscale secret image with resolution m ₁ ×n ₁ Generate l ₀ pairs of random integer coordinate sequences T ₀ in the range of m ₀ ×n ₀ from the key k ₁ , where Generate m ₀ ×n ₀ random integer matrices in the range [0,l-1] from the key k ₂ Where m ₀ =m ₂ /(2r-1), n ₀ =n ₂ /(2r-1);

Step 5: Read the integer coordinates (x, y) in T ₀ sequentially, calculate the coordinate position of (x, y) in M, and intercept the mosaic image M of (2r-1)×(2r-1) from M _t ;

Step 6: Correct the rotation of M _t to M′ _t and satisfy that the center of mass of M _t is in the direction agreed by the center of M′ _t , and then find the sum from all H″ _k , k=0,1,…,l-1 M _t ′ is the closest corresponding number Will Convert to secret value X _Cur ;

The 7th step: calculate random angle of rotation α _ind ∈ [0,2π) by the 6th step same method that claim 1 provides, M _t is rotated counterclockwise-α _ind to obtain M "_t;

Step 8: Convert X _Cur to log ₂ l-digit binary numbers and add them to the binary sequence B, and calculate M″ _t and The pixel point difference d, if d<T, the secret information X _Cur is not destroyed, then the binary number composed of log ₂ l bits 1 is added to B′, otherwise, log ₂ l bits 0 are formed The binary number of is added to B';

Step 9: Repeat steps 5 to 8 until all integer coordinates (x, y) in T ₀ are read;

Step 10: Convert B and B′ to a decimal number with P ₁ binary digits as a group, and then organize it into a P ₁ -level grayscale secret image of m ₁ ×n ₁ and the authentication image and output.

Preferably, step 2 maps ind _k to a random rotation angle α _ind between [0,2π), and then converts all H _k into circular images with random rotation angle α _ind and radius r The specific method is formula (1), where α _ind = ind _k · 2π/n:

H′ _k =Rot(H _k ,r,α _ind ) (1);

The third step is to correct all H′ _k rotations to H″ _k and satisfy the H″ _k centroid in the direction agreed by the center of H″ _k . The specific method is to correct all H′ _k rotations according to formula (12) to H″ _k and Satisfied that the H″ _k centroid is in the horizontal direction on the right side of the H″ _k circle center:

H″ _k = Rot _mass (H′ _k ) (12);

Step 6. The specific method to correct the rotation of M _t to M′ _t and satisfy the centroid of M′ _t in the direction agreed by the center of M′ _t is to correct the rotation to M′ _t according to formula (13) and satisfy the centroid of M′ _t to be in the direction of M′ t. ′ _t in the horizontal direction on the right side of the center of the circle:

M′ _t = Rot _mass (M _t ) (13);

In formula (12) and formula (13), function Rot _mass () is centroid rotation function, it is characterized in that comprising the following steps:

Step 3.1: Remember Calculate the centroid of H′ _k (x _mass , y _mass ):

Step 3.2: Calculate the geometric inclination angle of the center of mass of H′ _k relative to the center of circle of H′ _k ;

Step 3.3: Rotate H′ _k counterclockwise Get H″ _k , and then output H″ _k ;

Preferably, the specific method for calculating the centroid (x _mass , y _mass ) of H′ _k in step 3.1 is formula (14):

The specific method for calculating the geometric inclination of the H′ _k centroid relative to the H′ _k circle center in step 3.2 is to calculate the geometric inclination angle of the H′ _k centroid relative to the H′ _k circle center according to formula (15)

Step 3.3 Rotate H′ _k counterclockwise The concrete method that obtains H " _k is formula (16):

Preferably, the step 5 calculates the coordinate position of (x, y) in M. The specific method of intercepting the mosaic image M _t of (2r-1)×(2r-1) from M is formula (17):

Step 6 Find the closest distance to M _t ′ from all H″ _k , k=0,1,…,l-1 corresponding number The specific method is formula (18):

In formula (18), is the secondary distance between M′ and H″ _k .

Preferably, step 6 will The specific method of converting to the secret value X _Cur is formula (19):

In the 7th step, M _t is rotated counterclockwise-α _ind to obtain M″ _t . The specific method is:

M″ _t = Rot(M _k ,r,-α _ind ) (20);

Step 8 Calculate M″ _t and The specific method of the difference d of the pixel point is formula (21):

Compared with prior art, the beneficial effect of the present invention is:

①The information hiding capacity of the traditional search-based carrierless information hiding method is extremely low, involving a large number of images and texts intensively transmitted in the channel, although a single image and text are unmodified images and texts, and a large number of images and texts in the channel Medium-intensive transmission will also arouse the suspicion of attackers, so the proposed strategy cannot resist steganalysis; generative carrierless information hiding based on texture synthesis can only produce texture images with simple textures, but cannot generate meaningful images; The traditional information hiding methods based on mosaic puzzles all involve the modification of the cover image, which will leave traces of modification in the carrier, making it difficult to resist steganalysis attacks.

Different from the traditional method, the present invention converts a plurality of grayscale images with a resolution of (2r-1)×(2r-1) into a circular image with a pseudo-random corner and a radius of r, by generating a sequence of random integer coordinates to Determine the hidden position of the bit string divided by the dense image on the cover, and disperse the mean error generated by placing the circular image to the surrounding circular image with dense bit embedding. In the process of embedding, it does not involve any modified embedding of transformation parameters, but only uses the rotation angle and placement position of the circular image to fully and effectively express the secret information bits. The embedding and non-embedding pixels of the bunker are circular images with geometric corners added, and have nothing to do with the initial corners, so no modification traces related to secret information will be left on the embedding carrier, thus hiding the secret Difficult to discover and thus resistant to steganalysis.

②The traditional mosaic-based information hiding method needs to embed the transformation parameters directly into the dense bunker. Due to the hidden capacity problem, the space-domain reversible information hiding method based on LSB or related to it is usually used. These embedding methods will not only leave modification traces on the embedding carrier, but at the same time, due to the modification of non-significant bits, the embedded parameters cannot be extracted when attacked, so that the secret information cannot be accurately reconstructed.

Different from the traditional information hiding method based on mosaic puzzles, the present invention expresses the binary bit string divided by the secret information through a circular image, uses centroid rotation image matching to restore the secret information, and uses the median filter to repair JPEG compression or noise For the abnormal pixels after the attack, each circular image only approximates 1 pixel on the cover image through the mean value, and neither JPEG compression nor random noise attack will have a large impact on the mean value of the circular image, so the circular image The represented secret information can be recovered accurately to a large extent, so it is more robust than traditional methods.

③The traditional mosaic-based secret image camouflage method does not consider the correctness of the extracted secret information due to the small embedding capacity, and the security considered is also very limited. The receiver reconstructs the secret information according to the secret cover received by the channel, but cannot effectively verify the authenticity of the reconstructed secret information.

However, the method of the present invention uses circular images with random corners to express different secret information at different positions, and is strictly dependent on the user key. Only by providing the correct key can the secret information be obtained and the extracted secret information Verified for accuracy, otherwise unavailable.

Description of drawings

Figure 1 is a flowchart of secret information embedding;

Fig. 2 is a flowchart of secret information extraction;

Fig. 3 is an embodiment, and the resolution is an 8-bit grayscale secret image of 64×64;

Fig. 4 is an embodiment, and the resolution is an 8-bit grayscale mask image of 128 * 128;

Fig. 5 a is an embodiment, 32 resolutions are 8-bit grayscale images of 65 * 65;

Fig. 5b is an embodiment, and Fig. 5a corresponds to 32 circular images after circular initialization and adding random corners;

Fig. 5c is an embodiment. After calculating the centroid in Fig. 5b, rotate the coordinate point of the centroid to the corresponding 32 circular images on the positive semi-axis of X;

Fig. 6 a is an embodiment, utilizes Fig. 5 b circular image, the partial image (Fig. 6 b resolution is higher, provides Fig. 6 b part by Fig. 6 a here by Fig. detail);

Figure 6b is an embodiment, using the circular image in Figure 5b to form an 8-bit dense grayscale mosaic image with a resolution of 8320×8320;

Fig. 7 is an embodiment, the secret image extracted by Fig. 6b according to the process of Fig. 2, relative to Fig. 3 PSNR = ∞;

Fig. 8 is an embodiment, by Fig. 6 b according to Fig. 2 flow process, proposes the authentication map that secret image is generated, is the blank image of all white, represents that all pixels of secret image all pass authentication;

Fig. 9a is an embodiment, carrying out the JPEG compression that quality factor is 50 to Fig. 6b, the partial image of the JPEG grayscale image that forms (Fig. 9b resolution is higher, provides Fig. 9b local details by Fig. 9a here);

Figure 9b is an embodiment, the JPEG grayscale image formed by performing JPEG compression with a quality factor of 50 on Figure 6b;

Fig. 10 is an embodiment, the secret image extracted by Fig. 9b according to the process of Fig. 2, relative to Fig. 3 PSNR = ∞;

Fig. 11 is an embodiment, the authentication map generated by extracting the secret image according to the process of Fig. 2 from Fig. 9b is a completely white blank image, indicating that all pixels of the secret image have passed the authentication;

Fig. 12a is an embodiment, carrying out the JPEG compression that quality factor is 80 to Fig. 6b, the partial image of the JPEG gray scale image that forms (Fig. 12b resolution is higher, provides Fig. 12b local details by Fig. 12a here);

Figure 12b is an embodiment, the JPEG grayscale image formed by performing JPEG compression with a quality factor of 80 on Figure 6b;

Fig. 13 is an embodiment, the secret image extracted by Fig. 12b according to the process of Fig. 2, relative to Fig. 3 PSNR = ∞;

Fig. 14 is an embodiment, the authentication map generated by extracting the secret image according to the process of Fig. 2 from Fig. 12b is a completely white blank image, indicating that all pixels of the secret image have passed the authentication;

Fig. 15a is an embodiment, the partial image of the noise-containing mosaic image formed after the strength of 8% noise attack is carried out on Fig. 6b (Fig. 15b has a higher resolution, here the local details of Fig. 15b are provided by Fig. 15a);

Fig. 15b is an embodiment, a noise-containing mosaic image formed after performing a noise attack with an intensity of 8% on Fig. 6b;

Fig. 16 is an embodiment, the secret image extracted by Fig. 15b according to the process of Fig. 2, relative to Fig. 3 PSNR = ∞;

Fig. 17 is an embodiment. The authentication map generated by extracting the secret image according to the process of Fig. 2 from Fig. 15b is a completely white blank image, indicating that all pixels of the secret image have passed the authentication;

Fig. 18a is an embodiment, a partial image of a mosaic image containing noise formed after a 20% noise attack is carried out on Fig. 6b (Fig. 18b has a higher resolution, here the local details of Fig. 18b are provided by Fig. 18a);

Fig. 18b is an embodiment, a noise-containing mosaic image formed after performing a noise attack with a strength of 20% on Fig. 6b;

Fig. 19 is an embodiment, the secret image extracted by Fig. 18b according to the process of Fig. 2, relative to Fig. 3 PSNR = ∞;

Fig. 20 is an embodiment. The authentication map generated by extracting the secret image according to the process of Fig. 2 from Fig. 18b is a completely white blank image, indicating that all pixels of the secret image have passed the authentication.

Detailed ways

Taking JAVA jdk1.8.0_20 as the case implementation environment, the embodiment of the present invention will be described in detail in conjunction with the accompanying drawings, but not limited to this implementation case, wherein Fig. 1 is a flow chart of secret information camouflage, and Fig. 2 is a flow chart of secret information recovery. Among them, the embedding-free camouflage method combining block rotation and mosaic includes the following steps:

Step 1: Input a P ₀ -level grayscale mask image with a resolution of m ₀ ×n ₀ P ₁ -level grayscale secret image with resolution m ₁ ×n ₁ And l P ₂ -level grayscale images with a resolution of (2r-1)×(2r-1) in pairs where l is an integer power of 2 and satisfies

For example: take a grayscale mask image T=(t _i,j ) _4×4 with a resolution of m ₀ ×n ₀ =4×4, t _i,j ∈{0,1,…,255}, the pixel value is {(10,20,30,40),(50,60,70,80),(90,100,110,120),(130,140,150,160)}, where (10,20,30,40),(50,60,70,80) , (90, 100, 110, 120), (130, 140, 150, 160) respectively correspond to the 0th, 1st, 2nd and 3rd lines of the grayscale mask image; the grayscale image with resolution m ₁ ×n ₁ = 2×2 is taken as gray Degree secret image S＝(s _i,j ) _2×2 , s _i,j ∈{0,1,…,255}, take l=32 pairwise grayscale images with a resolution of 65×65 where r=33, P ₀ , P ₁ , P ₂ =8,

Step 2: Generate a random integer ind _k ∈ [0,n-1] from the key k ₀ , k=0,1,...,l-1 and n is an integer greater than 0;

For example: taking the key k ₀ =9, taking n=8 can generate 32 random integers in the range of [0,7] (5,2,6,6,6,6,4,7,7,1,3 ,2,4,4,4,0,3,0,7,7,2,5,0,3,3,5,2,6,5,6,1,5);

Step 3: Map ind _k to a random rotation angle α _ind between [0,2π) according to formula (1), and then convert all H _k into circular images with random rotation angle α _ind and radius r

H' _k = Rot(H _k , r, α _ind ), α _ind = ind _k · 2π/n (1);

In formula (1), Rot() is a counterclockwise rotation function, H _k corresponds to the input original image, r is the radius of the output ring image, and α _ind is the counterclockwise rotation angle. The specific implementation steps of the Rot() function are shown in Section 3.1 Steps to 3.4 show:

Step 3.1: Initialize H′ _k according to formula (2)

Step 3.2: Pixels within H _k radius r Convert to polar coordinates according to formula (3) Will Rotate the angle α _ind counterclockwise and transform it into Cartesian coordinates (i ₁ , j ₁ ) according to formula (4), then Assign to the elements in H′ _k Wherein when i=r-1, j=r-1, then directly as

In formula (4), “[]” is a rounding operator;

Step 3.3: After each Find within the centered 5x5 field The pixels of which are not -1 are recorded as P ₀ , P ₁ ,…,P _count-1 , and the corresponding and The secondary distances of are recorded as d ₀ ,d ₁ ,…,d _count-1 in turn, and then use formula (5) to Perform interpolation fitting;

Step 3.4: output H′ _k ;

For example: take k=0, assume that the image corresponding to H ₀ is Figure 5a(0), take ind ₀ =5, n=8, and calculate the corresponding random rotation angle α _ind =ind ₀ ×2π/n according to formula (1) =5·2π/8≈3.92699

① According to step 3.1, initialize 5a(0), and the coordinates of the center point of the image will be is the center of the circle, the pixels within the range of radius r=33 are initialized to -1, and the rest of the pixels are initialized to 0;

②According to step 3.2, the pixels within the range of radius r=33 Convert to polar coordinates (ρ, θ) according to formula (3), where θ=arctan((33-i-1)/(j-33+1)), for example, the polar coordinates corresponding to the pixel point (0,32) are new coordinates after rotation Copy the pixel value of the coordinate position (0,32) in the image of Figure 5a (0) to the position corresponding to the coordinate point (32,3) in the image of Figure 5b (0);

③According to step 3.3, find the value of the pixel point of -1 in Figure 5b(0), assuming that the pixel value of the coordinate point (32,32) is -1, and search for the 5×5 area centered on (32,32) Find the pixel points whose pixel value is not equal to -1, use count to count, and calculate the distance from the center point, and use the weighted average method to interpolate and fit the center pixel point (32,32).

Step 4: Scan S as a sequence of 2-valued bits Generate l ₀ random integer coordinate sequences in the range of m ₀ ×n ₀ from the key k ₁ in, is the ith random integer coordinate, namely

For example, the 4 pixels of the grayscale secret image S with a resolution of 2×2 are {1,2,3,4}, and S is scanned as a binary bit sequence B=(1,0,0,0,0 ,0,0,0,0,1,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,0,1,0,0,0 ,0,0,), where B contains a total of 32 elements, and the key k ₁ =5 can generate l ₀ =7 random integer coordinate sequences T ₀ =((2 ,3),(0,2),(0,1),(2,1),(1,3),(3,3),(3,1));

Step 5: Initialize the P ₀ -level grayscale mask image with a resolution of m ₂ ×n ₂ And satisfy m ₂ =m ₀ (2r-1), n ₂ =n ₀ (2r-1), generate m ₀ n ₀ random integer matrices in the range of [0,l-1] from key k ₂ And r _{i, j} ∈ [0, l-1];

For example: take m ₂ =m ₀ (2r-1)=4·65=260, n ₂ =n ₀ (2r-1)=4·65=260, the grayscale mask image with a resolution of 260×260 can be initialized , generate 4×4 random integer matrices R={(23,5,2,19),(15,30,14,26) in the range of [0,l-1=31] from key k ₂ =5 ,(14,13,22,24),(8,11,24,26)}, where (23,5,2,19),…,(8,11,24,26) correspond to the R matrix Line 0 to line 3;

Step 6: For Intercept 2-value bit B _cur from B according to formula (6), convert it to an integer X _Cur , and further map X _Cur to an integer in the range of [0,l-1] according to formula (7) Then select the first from H′ _k , k=0,1,…,l-1 images And calculate the integer ind in the range of [0,n-1] according to formula (8), then press formula (1) to Rotate α _ind counterclockwise to get where α _ind =ind·2π/n, the Rotate counterclockwise to Place it on M with (x · (2r-1), y · (2r-1)) as the starting point of the upper left corner, (2r-1) × (2r-1) matrix small block;

B _cur = Trim(B, Cur·log ₂ l, log ₂ l) (6);

In formula (6), Trim() is a binary sequence truncation function, where B is the input binary bit sequence, the second parameter is the starting index position of the interception, and the third parameter corresponds to the interception length;

ind = (X _Cur + x y + y r _{x, y} + x r _{x, y} ) mod n (8);

H′ _k =Rot(H _k ,r,α _ind ) (1);

For example: taking (x, y)=(2,3) as an example, since (2,3)∈T ₀ , according to formula (6), five binary bits B _Cur =(1,0, 0,0,0) where Cur=0 is taken, and B _Cur is converted into a decimal number X _Cur , and X _Cur =1·2 ⁰ +0·2 ¹ +0·2 ² +0·2 ³ =1 is calculated, and Further map X _Cur by formula (7) as which is Where r _2,3 = 24; take n = 8, from formula (8) can get ind = (1+2×3+2×24+3×24) mod8=7, thus can be That is, Fig. 5b(25) is rotated counterclockwise by an angle of α ₇ according to formula (1) to obtain Then place it on the matrix block starting from (2,3) in M, where α ₇ =7×π×0.25=5.49779;

Step 7: Calculate according to formula (10) and the difference Δ _{x, y between t x,y} on the gray-scale mask image T, and then transfer Δ _x,y to the unprocessed pixels around t _x,y according to formula (11)

In formula (10), mean() is the mean value calculation function of the circular image block, which calculates the mean value of the pixels within the circular image range with the center of the input image as the center and the radius r

For example: Assume that the coordinate position (0,1) on M is to be placed Pick That is, corresponding to Figure 5b(5), assuming that the center of Figure 5b(5) is the center of the circle, and the mean value of all pixels within the radius r=33 is 27, then it is calculated according to formula (10) and (0,1) in T The difference Δ _0,1 =27-20=7 of the pixel point t _0,1 =20, according to formula (11), transfer the difference 7 to the surrounding unprocessed pixels t _1,0 ,t _1,1 ,t _1,2 where The reason why it does not spread to t _0,2 here is that (0,2)∈T ₀ =((2,3),(0,2),(0,1),(2,1),(1, 3), (3,3), (3,1)).

Step 8: Repeat steps 6 to 7 until all coordinate positions in T ₀ are processed;

Step 9: For Find the closest small block to pixel t _x,y on T from H′ _k ,k=0,1,…,l-1 according to formula (9) Calculate the integer ind in the range of [0, n-1] according to formula (8), and then use formula (1) to Rotate α _ind counterclockwise to get where α _ind =ind·2π/n, X _Cur =0, place it on M with (x·(2r-1), y·(2r-1)) as the starting point of the upper left corner, (2r-1)× On the matrix block of (2r-1);

In formula (9), mean() is a circular image block mean value calculation function, which calculates the mean value of pixels within a circular image range whose radius is r with the center of the input image as the center;

For example: Take (x,y)=(0,0) as an example, because Assuming that according to the formula (9), the small block closest to the pixel t _0,0 on T is found from H′ _k ,k=0,1,…,l-1 is H′ ₁ , and the image corresponding to H′ ₁ is shown in the figure As shown in 5b(1), take X _Cur =0, by R={(23,5,2,19),(15,30,14,26),(14,13,22,24),(8, 11,24,26)} Knowing that r _0,0 = 23, then according to formula (8), it can be calculated that ind = (0+0×0+0×23+0×23) mod8=0, thus α _ind =ind· 2π/n=0, then according to formula (1), Figure 5b(1) can be rotated counterclockwise by 0, and placed in M with (0·(2r-1),0·(2r-1))=( 0,0) is the matrix block starting from the upper left corner.

Step 10: Calculate The difference between t _x,y and Δ _x,y , then pass Δ _x,y to the unprocessed pixels around t _x,y

For example: Assume that the coordinate position (0,0) on M is to be placed Pick That is, corresponding to Figure 5b(1), assuming that the center of Figure 5b(1) is the center of the circle, and the mean value of all pixels within the range of radius r=33 is 36, then it is calculated according to formula (10) and on (0,0) in T The difference Δ _0,0 =36-10=26 of the pixel point t _0,0 =10, according to formula (11), transfer the difference 26 to the surrounding unprocessed pixels t _1,0 ,t _1,1 , where The reason why it does not spread to t _0,1 and t ₀ ,- ₁ here is that (0,1)∈T ₀ =((2,3),(0,2),(0,1),(2, 1), (1,3), (3,3), (3,1)) and pixel t _{0, -1} does not exist.

Step 11: Repeat steps 9 to 10 until all coordinate positions that do not belong to T ₀ are processed;

Perform the above steps to finally generate an embedded mosaic image.

The specific implementation steps of the corresponding secret image recovery method are as follows:

Step 1: Input a P ₀ -level grayscale mosaic image with a resolution of m ₂ ×n ₂ l P ₂ -level grayscale images with a resolution of (2r-1)×(2r-1) in pairs Secret image resolution m ₁ ×n ₁ , secret image gray scale P ₁ , initialize binary bit sequence B=Φ and B′=Φ, input key k ₀ , k ₁ , k ₂ and set threshold parameter T ,T>0;

For example: the input resolution is m ₂ ×n ₂ =260×260 8-level gray mosaic image M=(m _i,j ) _260×260 , 32 resolutions are (2r-1)×(2r-1) =(2×33-1)×(2×33-1)=65×65 8-level grayscale images of different pairs As shown in Fig. 5a(0) ~ Fig. 5a(31), here r = 33, the resolution of the secret image is 2×2, the gray scale of the secret image is 8, and the binary bit sequence B=Φ and B′= Φ, set key k ₀ =9, k ₁ =5, k ₂ =5 and set threshold parameter T=70;

Step 2: Generate a random integer ind _k ∈ [0,n-1] from the key k ₀ , k=0,1,...,l-1 and n is an integer greater than 0, according to formula (1) will ind _k is mapped to a random rotation angle α _ind between [0,2π), and then all H _k are transformed into a circular image with a random rotation angle α _ind and radius r

For example: take the key k ₀ =9, generate 32 random integers in the range of [0,7] (5,2,6,6,6,6,4,7,7,1,3,2,4, 4,4,0,3,0,7,7,2,5,0,3,3,5,2,6,5,6,1,5), assuming n=8, k=0, then consisting of random integers (5,2,6,6,6,6,4,7,7,1,3,2,4,4,4,0,3,0,7,7,2,5,0, 3, 3, 5, 2, 6, 5, 6, 1, 5) we know: ind ₀ = 5, according to formula (1), ind ₀ is mapped to a random rotation angle between [0, 2π) α ₅ = 5· 2π/8＝3.9269909, then the That is, Figure 5a(0), transformed into a circular image with random rotation angle α ₅ =3.9269909 and radius r=33 That is, Figure 5b(0);

Step 3: For all H′ _k , the rotation is corrected to H″ _k according to formula (12) and the center of mass of H″ _k is in the direction agreed by the center of H″ _k circle;

H″ _k = Rot _mass (H′ _k ) (12);

In formula (12), the function Rot _mass () is the centroid rotation function, and the specific execution function is as follows:

Step 3.1: Remember Calculate the centroid (x _mass , y _mass ) of H′ _k according to formula (14):

Step 3.2: Calculate the geometric inclination angle of H′ _k centroid relative to H′ _k circle center according to formula (15)

Step 3.3: Rotate H′ _k counterclockwise according to formula (16) Get H″ _k , and then output H″ _k ;

For example: take k=0, can be That is, Fig. 5b(0) is rotated and corrected to Fig. 5c(0) according to formula (12) and satisfies that the centroid of Fig. 5c(0) is in the horizontal direction on the right side of the center of Fig. 5c(0). The specific implementation steps are as follows:

First calculate the centroid (x _mass .y _mass ) of H′ _k=0 according to formula (14), that is, the centroid (x _mass .y _mass ) of Figure 5b(0),

Then calculate the geometric inclination angle of the center of mass in Figure 5b(0) relative to the center of circle in Figure 5b(0) according to formula (15) Finally, rotate Figure 5b(0) counterclockwise according to formula (16) Get Figure 5c(0), and then output Figure 5c(0);

Step 4: Initialize the P ₁ -level grayscale secret image with resolution m ₁ ×n ₁ Generate l ₀ pairs of random integer coordinate sequences T ₀ in the range of m ₀ ×n ₀ from the key k ₁ , where Generate m ₀ n ₀ random integer matrices in the range [0,l-1] from the key k ₂ Where m ₀ =m ₂ /(2r-1), n ₀ =n ₂ /(2r-1);

For example: P ₁ =8-level grayscale secret image S=(s _i,j =0) _2×2 with initial resolution of m1×n1=2×2, l ₀ =7 generated by key k ₁ =5 Random integer coordinate sequence T 0 in the range of m ₀ ×n ₀ = 4 × 4 unequal in pairs T ₀ =((2,3),(0,2),(0,1),(2,1),( 1,3), (3,3), (3,1)), m ₀ n _{0 =} 16 random integer matrix R=((23, 5,2,19),(15,30,14,26),(14,13,22,24),(8,11,24,26)), where (23,5,2,19),… , (8,11,24,26) correspond to the 0th row to the 3rd row of the R matrix;

Step 5: Read the integer coordinates (x, y) in T ₀ sequentially, calculate the coordinate position (u, v) of (x, y) in M according to formula (17), and then use (u, v) as the starting point , intercept (2r-1)×(2r-1) mosaic image M _t from M;

For example: first read T ₀ = ((2,3),(0,2),(0,1),(2,1),(1,3),(3,3),(3,1) ), calculate the coordinate position (u=130, v=195) of (2,3) in M according to formula (17), where u=x·(2r-1)=2 ×(2×33-1)=130, v=y·(2r-1)=3×(2×33-1)=195, then take (130,195) as the starting point, intercept (2r-1) from M ×(2r-1)=(2×33-1)×(2×33-1)=65×65 mosaic image M ₀ ;

Step 6: Substituting M _t into formula (12), geometrically correcting to M′ _t according to formula (12) and satisfying that the center of mass of M _t is in the direction agreed by the center of M′ _t , and then according to formula (18) from all H″ _k ,k=0,1,…,l-1 to find the closest distance corresponding number According to formula (19) will Convert to secret value X _Cur ;

In formula (18), is the 2nd distance between M′ and H″ _k ;

For example, M ₀ is substituted into formula (12), geometrically corrected to M′ ₀ according to formula (12), and then according to formula (18) from all H″ _k , k=0,1,...,l-1, and then from Fig. Find the closest distance in 5c The corresponding number, assuming the found number Since the coordinates (2,3) in step 5 belong to T ₀ =((2,3),(0,2),(0,1),(2,1),(1,3),(3, 3), the first coordinate in (3,1)), so r _2,3 corresponds to the random integer matrix R=((23,5,2,19),(15,30,14,26),(14 ,13,22,24),(8,11,24,26)), the first value 23 in (8,11,24,26)), so r ₂ , ₃ =23; set r _2,3 =23, l=32, x=2, y=3 are substituted into formula (19), and can be calculated as

The 7th step: by the 6th step same method that claim 1 provides, calculate the integer ind in [0,n-1] scope by formula (8), M _t is rotated counterclockwise by formula (20)-α _ind Thus M″ _t is obtained, where α _ind =ind·2π/n

M″ _t = Rot(M _k ,r,-α _ind ) (20);

For example: calculate the integer ind=(X _Cur +x·y+y·r _x,y +x·r _x,y ) mod n=(1+2×3) in the range of [0,7] according to formula (8) +3×23+2×23) mod8=7, rotate M ₀ counterclockwise according to formula (20)-α _ind , where α _ind =7×π×0.25=5.49779, thereby obtaining M″ ₀ ;

Step 8: Convert X _Cur to log ₂ l-digit binary numbers and add them to the binary sequence B, then calculate M″ _t and The pixel point difference d, if d<T, the secret information X _Cur is not destroyed, then the binary number composed of log ₂ l bits 1 is added to B′, otherwise, log ₂ l bits 0 are formed The binary number of is added to B';

For example: X _Cur =1 is stored in B in binary form (10000) ₂ order, calculate M " ₀ and the difference d=29＜T=70 of the pixel point of Fig. 5b (25) by formula (21), To prove that the secret information X _Cur of this coordinate point has not been destroyed, add the binary number (11111) ₂ composed of 5 digits 1 to B′, otherwise, add the binary number (00000) ₂ composed of 5 digits 0 into B';

Step 9: Repeat steps 5 to 8 until all integer coordinates (x, y) in T ₀ are read;

Step 10: Convert B and B′ to a decimal number with P ₁ binary digits as a group, and then organize it into a P ₁ -level grayscale secret image of m ₁ ×n ₁ and the authentication image and output.

For example: Assuming that the final B=(10000000010000001100000000100000000) ₂ is converted into a decimal number (1,2,3,4) with 8 binary digits as a group, B'=(111111111111111111111111111111111) ₂ , with 8 digits One group of binary digits is converted into a decimal number (255,255,255,255), and then organized into a 2×2 8-level grayscale secret image S=( _si,j ) _2×2 and an authentication image A=(a _{i, j} ) _2×2 and output.

Fig. 3 is an embodiment, and the resolution is an 8-bit grayscale secret image of 64×64;

Fig. 4 is an embodiment, and the resolution is an 8-bit grayscale mask image of 128 * 128;

Fig. 5 a is an embodiment, 32 resolutions are 8-bit grayscale images of 65 * 65;

Fig. 5b is an embodiment, and Fig. 5a corresponds to 32 circular images after circular initialization and adding random corners;

Fig. 5c is an embodiment. After calculating the centroid in Fig. 5b, rotate the coordinate point of the centroid to the corresponding 32 circular images on the positive semi-axis of X;

Fig. 6 a is an embodiment, utilizes Fig. 5 b circular image, the partial image (Fig. 6 b resolution is higher, provides Fig. 6 b part by Fig. 6 a here by Fig. detail);

Figure 6b is an embodiment, using the circular image in Figure 5b to form an 8-bit dense grayscale mosaic image with a resolution of 8320×8320;

Fig. 7 is an embodiment, the secret image extracted by Fig. 6b according to the process of Fig. 2, relative to Fig. 3 PSNR = ∞;

Fig. 8 is an embodiment, by Fig. 6 b according to Fig. 2 flow process, proposes the authentication map that secret image is generated, is the blank image of all white, represents that all pixels of secret image all pass authentication;

Fig. 9a is an embodiment, carrying out the JPEG compression that quality factor is 50 to Fig. 6b, the partial image of the JPEG grayscale image that forms (Fig. 9b resolution is higher, provides Fig. 9b local details by Fig. 9a here);

Figure 9b is an embodiment, the JPEG grayscale image formed by performing JPEG compression with a quality factor of 50 on Figure 6b;

Fig. 10 is an embodiment, the secret image extracted by Fig. 9b according to the process of Fig. 2, relative to Fig. 3 PSNR = ∞;

Fig. 11 is an embodiment, the authentication map generated by extracting the secret image according to the process of Fig. 2 from Fig. 9b is a completely white blank image, indicating that all pixels of the secret image have passed the authentication;

Fig. 12a is an embodiment, carrying out the JPEG compression that quality factor is 80 to Fig. 6b, the partial image of the JPEG gray scale image that forms (Fig. 12b resolution is higher, provides Fig. 12b local details by Fig. 12a here);

Figure 12b is an embodiment, the JPEG grayscale image formed by performing JPEG compression with a quality factor of 80 on Figure 6b;

Fig. 13 is an embodiment, the secret image extracted by Fig. 12b according to the process of Fig. 2, relative to Fig. 3 PSNR = ∞;

Fig. 14 is an embodiment, the authentication map generated by extracting the secret image according to the process of Fig. 2 from Fig. 12b is a completely white blank image, indicating that all pixels of the secret image have passed the authentication;

Fig. 15a is an embodiment, the partial image of the noise-containing mosaic image formed after the strength of 8% noise attack is carried out on Fig. 6b (Fig. 15b has a higher resolution, here the local details of Fig. 15b are provided by Fig. 15a);

Fig. 15b is an embodiment, a noise-containing mosaic image formed after performing a noise attack with an intensity of 8% on Fig. 6b;

Fig. 16 is an embodiment, the secret image extracted by Fig. 15b according to the process of Fig. 2, relative to Fig. 3 PSNR = ∞;

Fig. 17 is an embodiment. The authentication map generated by extracting the secret image according to the process of Fig. 2 from Fig. 15b is a completely white blank image, indicating that all pixels of the secret image have passed the authentication;

Fig. 18a is an embodiment, a partial image of a mosaic image containing noise formed after a 20% noise attack is carried out on Fig. 6b (Fig. 18b has a higher resolution, here the local details of Fig. 18b are provided by Fig. 18a);

Fig. 18b is an embodiment, a noise-containing mosaic image formed after performing a noise attack with a strength of 20% on Fig. 6b;

Fig. 19 is an embodiment, the secret image extracted by Fig. 18b according to the process of Fig. 2, relative to Fig. 3 PSNR = ∞;

Fig. 20 is an embodiment. The authentication map generated by extracting the secret image according to the process of Fig. 2 from Fig. 18b is a completely white blank image, indicating that all pixels of the secret image have passed the authentication.

Claims (10)

1. A non-embedding camouflage method combining block rotation and mosaic is characterized by comprising the following steps:

step 1: input resolution of m₀×n₀P of₀Gray scale mask imageResolution of m₁×n₁P of₁Gray scale secret imageAnd l P with resolution of (2r-1) × (2r-1) which are unequal two by two₂Gray scale imageWherein, P₀,P₁Is an integer of 0 or more, r is a positive integer of 0 or more, l is an integer power of 2 and satisfies m₀n₀＞l₀,

Step 2: by a key k₀Generating a random integer ind_k∈[0,n-1]K is 0,1, …, l-1, and n is an integer greater than 0;

and 3, step 3: will ind_kMapping to random rotation angle α between [0,2 π))_indThen all H are added_kConversion to random corner α_indAnd a circular image of radius r

And 4, step 4: scanning S into a 2-valued bit sequenceBy a key k₁Generation of l₀M are different in pairs₀×n₀Random integer coordinate sequence within rangeWherein,is the ith random integer coordinate, i.e.

And 5, step 5: initializing resolution as m₂×n₂P of₀Gray scale mask imageAnd satisfy m₂＝m₀(2r-1),n₂＝n₀(2r-1) from the secret key k₂Generating m₀n₀A (0, l-1)]Random integer matrix within rangeAnd r is_i,j∈[0,l-1]；

And 6, step 6: for theTruncating 2-valued bits B from B_curConvert it to an integer X_CurAnd further converting X_CurThe mapping is [0, l-1 ]]Integer within the rangeThen from H'_kK is 0,1, …, l-1Sheet imageAnd calculateRandom corner α_indWill beRotate counterclockwise toPlacing it on a matrix patch of size (2r-1) × (2r-1) chosen on M starting from (x (2r-1), y (2r-1)) at the top left corner;

and 7, step 7: computingAnd a pixel T on the gray mask image T_x,yDifference Δ therebetween_x,yThen will be_x,yIs transmitted to t_x,ySurrounding unprocessed pixels

And 8, step 8: repeatedly executing the steps 6 to 7 until the processing T is finished₀All coordinate positions in (1);

step 9: for theFrom H'_kK is 0,1, …, l-1, and T is found_x,yClosest small blockAnd calculateRandom corner α_indWill beRotate counterclockwise toPlacing it on a matrix patch of size (2r-1) × (2r-1) chosen on M starting from (x (2r-1), y (2r-1)) at the top left corner;

step 10: computingAnd t_x,yDifference Δ therebetween_x,yThen will be_x,yIs transmitted to t_x,ySurrounding unprocessed pixels

Step 11: repeatedly executing the 9 th step to the 10 th step until the treatment is finished and does not belong to T₀All coordinate positions of (a);

step 12: and outputting the final M as the mosaic image after the mosaic is embedded and encrypted.

2. The masquerading method without embedding of combining block rotation and mosaic as claimed in claim 1, wherein in step 3, ind_kMapping to random rotation angle α between [0,2 π))_indThen all H are added_kConversion to random corner α_indAnd a circular image of radius rThe specific method comprises the following steps:

all H are expressed according to formula (1)_kConversion to random corner α_indAnd a circular image of radius rWherein, α_ind＝ind_k·2π/n；

H′_k＝Rot(H_k,r,α_ind) (1)；

In the formula (1), Rot () is a counterclockwise rotation function, H_kCorresponding to the input original image, r is the output annular image radius, α_indIs a counterclockwise rotation angle.

3. The embeddable masquerading method combining block rotation and mosaic as claimed in claim 2, wherein the Rot () function in equation (1) is implemented as shown in steps 3.1 to 3.4:

step 3.1: is represented by formula (2) to H'_kAnd (3) initializing:

step 3.2: h is to be_kPixels within radius rConversion into polar coordinates by equation (3)Will be provided withRotate counterclockwise α_indThe angle is converted into rectangular coordinate (i) according to formula (4)₁,j₁) Then will beValue to H'_kElement (1) ofWherein, when i is r-1, j is r-1, then directly willAs

In the formula (4), [ ] is 4 or 5 in the operator;

step 3.3: at each one atCentered 5 x 5 in-field searchThe pixels of (2) are divided into pixels which are not-1The point is denoted as P₀,P₁,…,P_count-1Corresponding thereto and2 times of distance are sequentially marked as d₀,d₁,…,d_count-1Then using the pair of formula (5)Carrying out interpolation fitting;

step 3.4: h'_kAnd (6) outputting.

4. The method of claim 1, wherein in step 6, 2-valued bits B are truncated from B_curThe specific method comprises the following steps:

intercept log from B according to equation (6)₂l 2-valued bits B_curIf the total number of truncations is less than log₂If there are l, intercepting all the remaining bits;

B_cur＝Trim(B,Cur·log₂l,log₂l) (6)；

in the formula (6), Trim () is a 2-value sequence truncation function, wherein the 1 st input parameter of the Trim () function is an input 2-value bit sequence, the 2 nd parameter is an interception start index position, the 3 rd parameter corresponds to an interception length, Cur is a currently intercepted 2-value bit index, and Cur is greater than or equal to 0;

in step 6, X is_CurThe mapping is [0, l-1 ]]Integer within the rangeThe specific method of (3) is formula (7):

in steps 6 and 9, random rotation angles α are calculated_indWill beRotate counterclockwise toThe specific method comprises the following steps:

calculating [0, n-1 ] according to equation (8)]An integer ind within the range, and then will be expressed by equation (1)Rotate counterclockwise α_indTo obtainWherein, α_ind＝ind·2π/n：

ind＝(X_Cur+x·y+y·r_x,y+x·r_x,y)modn (8)；

H′_k＝Rot(H_k,r,α_ind) (1)。

5. The embeddable masquerading method of combining block rotation and mosaic as claimed in claim 1, wherein in step 9, forFrom H'_kK is 0,1, …, l-1, and T is found_x,yClosest small blockThe specific method comprises the following steps:

from H 'according to formula (9)'_kK is 0,1, …, l-1, and T is found_x,ySmall block with closest mean

In the formula (9), mean () is a calculation function of the mean value of the circular image block, and the calculated mean value of pixels in the range of the circular image with the center of the input image as the center of a circle and the radius of r is used as the mean value of the pixels;

in step 10, calculateAnd t_x,yDifference Δ therebetween_x,yThe specific method of (3) is formula (10):

in step 10,. DELTA._x,yIs transmitted to t_x,ySurrounding unprocessed pixelsThe specific method of (3) is formula (11):

6. a combined block rotation and mosaic embeddable recovery method as claimed in claim 1, comprising the steps of:

step 1: input resolution of m₂×n₂P of₀Gray scale mosaic imagel P with resolution of (2r-1) × (2r-1) different pairwise₂Gray scale imageSecret image resolution m₁×n₁The gray scale P of the secret image₁Initializing binary bit sequence B phi and B' phi, and inputting key k₀,k₁,k₂Setting a threshold parameter T, wherein T is more than 0;

step 2: by a key k₀Generating a random integer ind_k∈[0,n-1]K is 0,1, …, l-1 and n is an integer greater than 0, and ind_kMapping to random rotation angle α between [0,2 π))_indThen all H are added_kConversion to random corner α_indAnd a circular image of radius r

And 3, step 3: to all H'_kThe rotation correction is H ″)_kAnd satisfy H ″)_kThe center of mass is H_kThe direction of the circle center is appointed;

and 4, step 4: initializing resolution as m₁×n₁P of₁Gray scale secret imageBy a key k₁Generation of l₀M are different in pairs₀×n₀Random integer coordinate sequence T within range₀WhereinBy a key k₂Generating m₀×n₀A (0, l-1)]Random integer matrix within rangeWherein m is₀＝m₂/(2r-1),n₀＝n₂/(2r-1)；

And 5, step 5: reading T in sequence₀The coordinate position of (x, y) in M is calculated, and the mosaic image M of (2r-1) × (2r-1) is cut out from M_t；

And 6, step 6: will M_tRotation correction is M'_tAnd satisfy M_tCenter of mass at M'_tIn the direction of the center convention, then from all H ″)_kK-0, 1, …, l-1 for a sum M'_tThe closest distanceCorresponding numberingWill be provided withConversion to secret value X_Cur；

Step 7 of calculating the random rotation angle α in the same manner as in step 6 given in claim 1_indE is [0,2 π ]), adding M_tRotate counterclockwise- α_indObtaining M_t；

And 8, step 8: mixing X_CurConversion to log₂Adding l bit 2 series number to 2 value sequence B, calculating M ″_tAndif d is less than T, secret information X_CurNot destroyed, log₂A 2-ary number consisting of l bits 1 is added to B', otherwise log₂A 2-system number consisting of l bits 0 is added to B';

step 9: repeatedly executing the 5 th step to the 8 th step until T₀Reading all the integer coordinates (x, y);

step 10: b and B' are substituted by P₁Each 1 set of 2 bits is converted to a 10 digit and then organized into m₁×n₁P of₁Gray scale secret imageAnd authenticating the imageAnd output.

7. A process as claimed in claim 6A non-embedding recovery method combining block rotation and mosaic is characterized in that step 2 is to divide ind_kMapping to random rotation angle α between [0,2 π))_indThen all H are added_kConversion to random corner α_indAnd a circular image of radius rIs of the formula (1), wherein α_ind＝ind_k·2π/n:

H′_k＝Rot(H_k,r,α_ind) (1)；

Step 3 to all H'_kThe rotation correction is H ″)_kAnd satisfy H ″)_kThe center of mass is H_kThe specific method in the direction of circle center convention is to all H'_kRotation corrected to H ″, as per equation (12)_kAnd satisfy H ″)_kThe center of mass is H_kIn the horizontal direction on the right side of the circle center:

H″_k＝Rot_mass(H′_k) (12)；

step 6 is to mix M_tRotation correction is M'_tAnd satisfy M'_tCenter of mass at M'_tThe specific method in the direction of circle center convention is that the rotation correction is M 'according to formula (13)'_tAnd satisfy M'_tCenter of mass at M'_tIn the horizontal direction on the right side of the circle center:

M′_t＝Rot_mass(M_t) (13)；

in the equations (12) and (13), the function Rot_mass() Is a centroid rotation function, and is characterized by comprising the following steps:

step 3.1: note the bookCalculate H'_kCenter of mass (x)_mass,y_mass)：

Step 3.2: calculate H'_kCenter of mass relative to H'_kThe geometric inclination of the circle center;

step 3.3: h'_kRotate counterclockwiseTo obtain H_kThen, H ″' is added_kAnd (6) outputting.

8. The method for recovering from mosaic and rotation of combined blocks according to claim 6, wherein H 'is calculated at step 3.1'_kCenter of mass (x)_mass,y_mass) The specific method of (3) is formula (14):

step 3.2 calculating H'_kCenter of mass relative to H'_kThe specific method of the geometric dip angle of the circle center is to calculate H 'according to the formula (15)'_kCenter of mass relative to H'_kGeometric inclination of the center of a circle

Step 3.3 is H'_kRotate counterclockwiseTo obtain H_kThe specific method of (3) is formula (16):

9. a method for non-embedding restoration combining block rotation and mosaic as claimed in claim 5, wherein the coordinate position of step 5 (x, y) in M is calculated by truncating (2r-1) × (2r-1) mosaic image M from M_tThe specific method of (1) is formula (17):

step 6 from all H ″)_kK-0, 1, …, l-1 for a sum M'_tThe closest distanceCorresponding numberingThe specific method of (3) is formula (18):

in the formula (18), the reaction mixture,are M' and H ″)_k2 times of distance.

10. The method of claim 5, wherein step 6 is to combine block rotation and mosaic for non-embedding recoveryConversion to secret value X_CurThe specific method of (3) is represented by formula (19):

step 7 is to mix M_tRotate counterclockwise- α_indObtaining M_tThe specific method comprises the following steps:

M″_t＝Rot(M_k,r,-α_ind) (20)；

step 8, calculate M ″)_tAndis formed by a plurality of pixelsA specific method of the point difference d is formula (21):