CN102103738B - Method for generating and authenticating digital image tampered content recoverable variable capacity watermarks - Google Patents

Abstract

A variable-capacity watermark generation and authentication method with recoverable digital image tampering content. A 2×2 image block is divided into a smooth image block and a non-smooth image block. The smooth image block extracts 6-bit features, and the non-smooth image block extracts 12 bits. Bit features, encrypting image block features to generate image block recovery watermarks and randomly embedding them into other image blocks, and judging its authenticity by comparing the consistency of image block features and combining block neighborhood characteristics. For image blocks judged to be tampered, different tampering recovery operations are performed in two steps according to whether the corresponding recovery watermark has been tampered with to improve the quality of tampering recovery. The present invention generates a variable-capacity recovery watermark according to the characteristics of the image block itself, and the recovery watermark is used for tampering location and tampering recovery at the same time, while ensuring sufficient image features, the watermark embedding capacity is reduced, and the quality of the watermarked image and the quality of tampering recovery are taken into account; at the same time It improves security and can resist known forgery attacks such as collage attack and mean attack.

Description

Generation and authentication method of variable capacity watermark with recoverable digital image tampering content

technical field

The invention relates to a generation and authentication method of an authentication watermark whose tampered content of a word image can be recovered, which is used for detecting the authenticity and integrity of a digital image and recovering the tampered image. the

Background technique

With the development of computer, digital imaging and network communication technology, digital image has become the main source for people to obtain and exchange information and one of the important carriers of information dissemination. The digital storage of images and the emergence of various image processing software make the editing, modification and synthesis of digital images very simple. Digital image processing technology has improved the display quality of images, but tampering and forgery of images, if used in news media, court evidence, scientific discoveries, etc., will bring serious damage to the integrity of the society, the credibility of the government, and the authenticity of science. Negative impact. Therefore, how to detect and identify the authenticity and integrity of digital images has become a frontier research topic at home and abroad in recent years. It not only has important academic value, but also has important social significance and broad application prospects. the

Restorable watermarking can not only detect and identify the authenticity and integrity of digital images, but also approximate the original content of tampered areas. "Tamper recovery" puts forward higher requirements for authentic digital watermarking, how to improve the quality of recovered images is an important goal of recoverable watermarking research. The digital image recoverable watermarking algorithm first divides the image into image blocks of the same size to realize tampering location, and the recovery watermark generated according to the characteristics of the image block is not embedded in the image block itself, but embedded in other image blocks based on the key. This processing reduces the possibility of the image block being tampered with while restoring the watermark, which is beneficial to the restoration of the tampered block, but makes tampering detection relatively difficult. The tamper localization performance of the recoverable watermarking algorithm directly affects the quality of tamper recovery, because: if the tamper is not detected, or the real image is misjudged as tampered, it will cause the tampering in the restored image to be unrecovered or the real image will be wrongly detected. recover. In order to solve the tamper localization problem of recoverable watermark, Lin et al. , 38(12):2519-2529) proposed to attach a 2-bit authentication watermark to each image block to detect the authenticity of the image block. However, since the authentication watermark is block-independent, the authentication watermark cannot resist collage attack. That is to say, assuming that the same key is used to generate two watermarked images Y1 and Y2 of the same size, and the corresponding image block in Y2 is replaced by the image block in Y1, the authentication watermark cannot detect the tampering. Even if the authentication watermarking bits of each image block are increased to 32 bits, the self-embedded watermarking algorithm with high-quality recovery ability proposed by Qian et al. -embedding with high-quaility restoration capability, Digital Signal Processing, 2011(21): 278-286), the additional authentication watermark still cannot resist collage attack. Moreover, the additional authentication watermark increases the watermark embedding capacity, and also degrades the quality of the watermarked image. the

In addition, the recovery watermark generation method is an important factor affecting the positioning accuracy, tamper recovery quality and security. There are two most commonly used methods for generating and restoring watermarks: ① Quantization and encoding of important DCT (discrete cosine transform) coefficients of 8×8 image blocks, the disadvantage of which is low positioning accuracy. ② The average value of 2×2 image blocks has high positioning accuracy, but it is vulnerable to constant average attack and the recovery quality of non-smooth image blocks is not good. In the algorithm for generating a restored watermark using the average value of 2×2 blocks, assuming two image blocks B1=(30, 30, 30, 30) and B2=(1, 1, 59, 59), their average values are the same, and both for 30. So their recovered watermarks are also the same. If block B2 is replaced by image block B1, the recovery watermarking algorithm that detects the authenticity of the block by comparing the recovered watermark cannot detect this tampering (i.e., cannot resist the constant-mean attack), and the algorithm that detects the authenticity of the block using an additional authentication watermark will not be good. recovery quality. This is because the corresponding restoration of the watermark can only reconstruct the mean feature of B2, but not the texture feature of B2. the

If the image block restoration watermark saves the texture feature of the image block in addition to the mean value feature of the image block, the restoration quality of the algorithm and the ability to resist the constant mean value attack will be improved. But saving the textural features of the image block will increase the watermark capacity, and adding texture features to smooth blocks like B1 is redundant. the

Contents of the invention

The present invention solves the technical problem, and the adopted technical solution is: a variable-capacity watermark generation and authentication method with recoverable content of digital image tampering, including the following steps:

A. Watermark generation and embedding

A1. Image segmentation and classification: Divide the original image X with a size of 2m×2n into m×n non-overlapping 2×2 image blocks X _i ={ _xi1 , _xi2 , _xi3 , _xi4 }, Wherein, i is the image block number, i=1, 2,..., N, N=m×n is the number of image blocks; and the image blocks are divided into smooth image blocks and non-smooth image blocks according to the content of the image blocks kind;

A2. Pseudo-random sequence and block chain generation: Generate a real-valued pseudo-random sequence of length N based on the user key Key R={r _i |i=1, 2,...,N}, ordered by the index of R Sequence generation block chain {(X _i ,X _i′ )|i, i′∈[1,N]}, i′ is the value of the i-th element in the R index ordered sequence, and X _i’ is the mapping of X _i piece;

A3. Variable capacity feature extraction: calculate and generate v-bit block feature F _i ={f _i1 , f _i2 , . . . , f _iv } for each image block X _i . Among them, f _i2 ~ f _i6 is the 5-bit binary code of the average value of the upper 5 bits of the image block _Xi , if the image block _Xi is a smooth image block, then v=6 and f _i1 =0, otherwise v=12, f _i1 =1, f _i7 ~ f _i12 are the detail coding of the non-smooth image block;

A4. Variable-capacity watermark generation: use the i-th random number r _i of the pseudo-random sequence R to generate a binary pseudo-random sequence B _i ={b _ij |j=1, 2,...12}, and encrypt image block features F _i = {f _i1 , f _i2 , ..., f _iv } generate restored watermark W _i = {w _i1 , w _i2 , ..., w _iv },

${\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; \&CirclePlus;</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}$

为异或操作；对平滑图像块b_ij的下标j取值范围为1～6，对非平滑图像块b_ij的下标j取值范围为1～12；  in,

is an XOR operation; the value range of subscript j for smooth image block b _ij is 1-6, and the value range of subscript j for non-smooth image block b _ij is 1-12;

A5 _. Variable-capacity watermark embedding: Embed the restored watermark W _i of the image block Xi into the least significant bit of its mapping block Xi _' to generate a watermarked image block Y _i' ,

If _Xi is a non-smooth image block,

If _Xi is a smooth image block,

B. Watermark extraction and comparison

再按A2步相同的操作和密钥Key生成随机序列R和块链 

B1. Blocking, pseudo-random sequence and block chain generation: Y ^* is the tested image received after the watermarked image Y is transmitted, and Y ^* is divided into 2×2 blocks according to step A1

Then follow the same operation and key Key in step A2 to generate random sequence R and block chain

按A3步操作计算出图像块特征 

同时根据从其映射图像块 

低有效位提取恢复水印 

重构出块特征 

B2. Block feature calculation and reconstruction: for each tested image block

Calculate the image block features according to the A3 step operation

while mapping image blocks according to

Least Significant Bit Extraction and Restoration of Watermark

Refactor block feature

${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; L</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; \&CirclePlus;</span>{\mathrm{<span\; class="notranslate">\; B</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}$

低位提取水印信息 

按以下公式得到，  Among them, B _i is a binary pseudo-random sequence of length 12 generated based on r _i , from which the image block is mapped

Low bit extraction watermark information

According to the following formula,

高位计算得到的块特征 

和从其映射块 

低有效位提取恢复水印重构的块特征 按下列公式生成比较矩D＝{d_i|i＝1，2，...，N}，  B3. Classification feature comparison: compare the tested image blocks

Block characteristics obtained by high-level calculation

and map blocks from it

Extraction of Least Significant Bits to Recover Block Features of Watermark Reconstruction Generate comparative moment D={d _i |i=1, 2,..., N} according to the following formula,

in $c=6(1+f_{i1}^{*});$

C. Tamper detection

C1. Calculate and generate the neighborhood characteristic matrix Δ={δ _i |i=1, 2, ..., N} of the comparison matrix D,

δ _i ＝∑d _j , j=i±1, i±n, i+n±1, in±1

C2. According to the comparison matrix D and its neighborhood feature matrix Δ, generate the initial state tampering detection matrix T ⁰ $T^0=\left(t_i^0\right|i=\mathrm{1,2},...,N),$

C3. Generate a tampering detection matrix T according to the initial state tampering detection matrix ^T0 generated in the C2 step,

是真实的；  in j=i±1, i±n, i±n±1. t _i =1 means the image block under test tampered, t _i =0 means the image block under test

It is true;

D. Tamper recovery

If the elements in T are not all 0, it indicates that there is a tampered image block in the tested image, then perform the following two steps in turn on the tested image to obtain the tampered recovery image Y ^R ;

如果其映射块Y_i’ ^*是真实的，用B2步重构块特征 对 

进行恢复，  D1. Feature recovery: For image blocks judged to be tampered

If its mapped block Y _i' ^* is real, reconstruct block features with B2 step right

to restore,

Among them, Γ ^-1 () is the inverse function of the feature encoding function of the image block; at the same time, a label matrix L is generated to mark the tampered block that has not been restored,

利用与其相邻12个像素中的有效像素均值修正图像块 

中的每一个像素的值。  D2. Neighborhood recovery: if the elements in L are not all 0, the image block corresponding to l _i =1

Correct the image block by using the mean value of the effective pixels in the 12 adjacent pixels

The value of each pixel in .

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

1. Variable feature capacity: the present invention extracts block features of unequal length based on image block content—6 bits for smooth image blocks and 12 bits for non-smooth image blocks, so as to adaptively generate variable-capacity restoration watermarks, ensuring high-quality restoration While tampering with image blocks and resisting constant mean attack, the extracted feature capacity is minimized. the

2. The capacity of the recovery watermark is variable: no redundant information is added when the recovery watermark is generated, two types of block features with different lengths (6 and 12 bits) are encrypted to generate a recovery watermark of the same length, and no redundant information is added when the recovery watermark is generated. and use different methods to embed the variable-capacity restored watermark into other image blocks. While fully retaining the features of the image block, the capacity of the embedded recovery watermark is minimized. At the same time, the algorithm does not need to add an additional authentication watermark, which further reduces the embedding capacity of the watermark, thereby improving the quality of the watermarked digital image. the

3. Restoration of watermarks for tamper detection: Consistency detection of image blocks utilizes all restoration watermarks of image blocks—6 bits for smooth image blocks and 12 bits for non-smooth image blocks. All restored watermarks participate in consistency detection, which can not only improve the ability of the algorithm to resist constant mean attack and collage attack, but also lay a foundation for further improving the tamper detection performance of the algorithm. the

4. High security: first encrypt the features of the image block based on the key to generate recovery watermark information, and then randomly embed it into other image blocks based on the key, and use the embedded recovery watermark information to detect the consistency of the image block and establish the image Random dependencies between blocks. At the same time, the variable-capacity restored watermark preserves as much feature information as possible of the image block. These characteristics make the algorithm resistant to known forgery attacks, such as collage attack, constant mean attack, etc., thus improving the security of the algorithm. the

5. High restoration quality: First, 12 bits are used to save the features of non-smooth image blocks with rich details, so as to preserve as many features of image blocks as possible, laying the foundation for improving the quality of tampering restoration. Second, high localization accuracy (2×2 pixels) and excellent tamper detection performance provide the possibility to improve the quality of tamper recovery. Finally, for different tampered image blocks in the algorithm, different tampering recovery operations are performed in two steps to further improve the quality of the tampering recovery image. the

The present invention will be further described below in conjunction with embodiment. the

Description of drawings

Fig. 1 is a block diagram of the steps of generating and embedding a variable capacity restoration watermark according to an embodiment of the present invention. the

Fig. 2 is a schematic diagram of non-smooth image block classification and subclass encoding according to an embodiment of the present invention. the

Fig. 3 is a performance comparison diagram between the present invention and two existing recoverable watermarking algorithms under general tampering. the

Fig. 4 is a performance comparison diagram between the present invention and two existing recoverable watermarking algorithms under the constant mean value attack. the

Fig. 5 is a performance comparison diagram between the present invention and two existing recoverable watermarking algorithms under collage attack. the

Detailed ways

Example

A variable-capacity watermark generation and authentication method with recoverable tampered content of a digital image, including four parts: watermark generation and embedding, watermark extraction and comparison, tampering detection and tampering recovery. the

A. Watermark generation and embedding

Fig. 1 is a block diagram of specific steps of watermark generation and embedding in the method of the embodiment of the present invention, including the following five steps:

A1. Image segmentation and classification: Divide the original image X with a size of 2m×2n into m×n non-overlapping 2×2 image blocks X _i ={ _xi1 , _xi2 , _xi3 , _xi4 }, Wherein, i is the image block number, i=1, 2,..., N, N=m×n is the number of image blocks; and the image blocks are divided into smooth image blocks and non-smooth image blocks according to the content of the image blocks kind;

Image block classification in this embodiment: for a 2×2 image block X _i ={x _i1 , x _i2 , x _i3 , x _i4 }, if the difference between the two largest pixels and the two smallest pixels in the image block does not exceed h( In this example, h=3), the image block is a smooth image block, otherwise it is a texture image block.

During implementation, the value of the difference h between the two largest pixels and the two smallest pixels used for image block classification usually ranges from 2 to 16. The smaller the value of h, the more non-smooth image blocks are considered, and the restoration quality of the image The higher the value is, the larger the corresponding embedded watermark capacity will be, which will reduce the quality of the watermarked image. Conversely, the more smooth image blocks, the smaller the capacity of the corresponding embedded watermark, the higher the quality of the watermarked image, but the lower the quality of the restored image. the

A2. Pseudo-random sequence and block chain generation: Generate a real-valued pseudo-random sequence of length N based on the user key Key R={r _i |i=1, 2,...,N}, ordered by the index of R Sequence generation block chain {(X _i ,X _i′ )|i, i′∈[1,N]}, i' is the value of the i-th element in the R index ordered sequence, and X _i' is the mapping of X _i piece;

A3. Variable capacity feature extraction: calculate and generate v-bit block feature F _i ={f _i1 , f _i2 , . . . , f _iv } for each image block X _i . Among them, f _i2 ~ f _i6 is the 5-bit binary code of the average value of the upper 5 bits of the image block _Xi , if the image block _Xi is a smooth image block, then v=6 and f _i1 =0, otherwise v=12, f _i1 =1, f _i7 ~ f _i12 are the detail coding of the non-smooth image block;

计算，其中(.)_B表示整数的二进制编码。  In this embodiment, the binary coding of the high 5-bit average value of the image block _Xi is according to the formula

Compute, where (.) _B represents the binary encoding of the integer.

其中j’(j＝1，2，3，4)为图像块四个像素的索引有序序列。  The detail codes f _i7 -f _i12 of the non-smooth image blocks are obtained by classifying the image blocks. According to the appearance position of the two largest pixels in the 2×2 image block, the non-smooth image block is divided into 6 categories, and the largest two pixels are located in the image block: f _i7 ~ f _i9 above are coded as 001, and f _i7 ~ f i9 below are coded as 001 f _i9 is coded as 010, f _i7 to f _i9 on the left are coded as 011, f _i7 to f i9 on the right are coded as 100, f _i7 to f _i9 are coded as 101 for left oblique (upper left and lower right), and f i7 to f _i9 for right oblique (upper right, lower left) f _i7 ˜f _i9 are coded as 110, as shown in FIG. 2 . f _i10 ~ f _i12 is the uniform quantized binary code of the difference between the sum of two maximum pixels and the sum of two minimum pixels, namely

Where j' (j=1, 2, 3, 4) is an ordered sequence of indexes of four pixels of the image block.

A4. Variable-capacity watermark generation: use the i-th random number r _i of the pseudo-random sequence R to generate a binary pseudo-random sequence B _i ={b _ij |j=1, 2,...12}, and encrypt image block features F _i = {f _i1 , f _i2 , ..., f _iv } generate restored watermark W _i = {w _i1 , w _i2 , ..., w _iv },

${\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; \&CirclePlus;</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}$

为异或操作。对平滑图像块，伪随机序列B_i＝{b_ij|j＝1，2，…12}仅用前6个比特；对非平滑图像块，伪随机序列B_i＝{b_ij|j＝1，2，…12}的12个比特都用。  in,

is an XOR operation. For a smooth image block, the pseudo-random sequence B _i ={b _ij |j=1, 2,...12} only uses the first 6 bits; for a non-smooth image block, the pseudo-random sequence B _i ={b _ij |j=1 , 2,...12} all 12 bits are used.

A5. Variable-capacity watermark embedding: Embed the restored watermark W _i of the image block Xi into the least significant bit of its mapping block Xi _' to generate a watermarked image _block Y _' ,

If _Xi is a non-smooth image block,

If _Xi is a smooth image patch,

B. Watermark extraction and comparison

B1. Blocking, pseudo-random sequence and block chain generation: Y ^* is the tested image received after the watermarked image Y is transmitted, and Y ^* is divided into 2×2 blocks according to step A1 Then follow the same operation and key Key in step A2 to generate random sequence R and block chain

按A3步操作计算出图像块特征 

同时根据从其映射图像块 

低有效位提取恢复水印 

重构出块特征 

B2. Block feature calculation and reconstruction: for each tested image block

Calculate the image block features according to the A3 step operation

while mapping image blocks according to

Least Significant Bit Extraction and Restoration of Watermark

Refactor block feature

${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; L</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; \&CirclePlus;</span>{\mathrm{<span\; class="notranslate">\; B</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}$

低位提取水印信息 

按以下公式得到，  Among them, B _i is a binary pseudo-random sequence of length 12 generated based on r _i , from which the image block is mapped

Low bit extraction watermark information

According to the following formula,

和从其映射块 

低有效位提取恢复水印重构的块特征 

按下列公式生成比较矩D＝{d_i|i＝1，2，...，N}，  B3. Classification feature comparison: compare the tested image blocks Block characteristics obtained by high-level calculation

and map blocks from it

Extraction of Least Significant Bits to Recover Block Features of Watermark Reconstruction

Generate comparative moment D={d _i |i=1, 2,..., N} according to the following formula,

也就是说，对平滑图像块，比较 与 

前6个比特是否相同；对非平滑图像块，比较 与 

所有12个比特是否相同。  in

That is, for smooth image blocks, comparing and

Whether the first 6 bits are the same; for non-smooth image blocks, compare and

Are all 12 bits the same.

C. Tamper detection

C1. Calculate and generate the neighborhood characteristic matrix Δ={δ _i |i=1, 2, ..., N} of the comparison matrix D,

δ _i ＝∑d _j , j=i±1, i±n, i+n±1, in±1

Among them, the generation method of D's neighborhood feature matrix Δ={δ _i |i=1, 2, ..., N} is an existing method, and the specific method is detailed in the literature He Hong-jie, Zhang Jia-shu, Chen Fen. A self-recovery fragile watermarking scheme for image authentication with superior localization, Science in China Series F-Information Sciences, 2008.51(10): 1487-1507.

C2. According to the comparison matrix D and its neighborhood feature matrix Δ, generate the initial state tampering detection matrix T ⁰ $T^0=\left(t_i^0\right|i=\mathrm{1,2},...,N),$

C3. Generate a tampering detection matrix T according to the initial state tampering detection matrix ^T0 generated in the previous step,

j＝i±1，i±n，i±n±1。t_i＝1表示被测图像块 

被篡改，t_i＝0表示被测图像块 

是真实的；  in

j=i±1, i±n, i±n±1. t _i =1 means the image block under test

tampered, t _i =0 means the image block under test

It is true;

D. Tamper recovery

If the elements in T are not all 0, it indicates that there is a tampered image block in the tested image, then perform the following two steps in turn on the tested image to obtain the tampered recovery image Y ^R ;

如果其映射块Y_i’ ^*是真实的，用B2步重构块特征 

对 

进行恢复，  D1. Feature recovery: For image blocks judged to be tampered

If its mapped block Y _i' ^* is real, reconstruct block features with B2 step

right

to restore,

Among them, Γ ^-1 () is the inverse function of the feature encoding function of the image block; at the same time, a label matrix L is generated to mark the tampered block that has not been restored,

利用与其相邻12个像素中的有效像素均值修正图像块 

中的每一个像素的值。  D2. Neighborhood recovery: if the elements in L are not all 0, the image block corresponding to l _i =1

Correct the image block by using the mean value of the effective pixels in the 12 adjacent pixels

The value of each pixel in .

Effect of the present invention can be verified by following performance analysis and experiments

When analyzing and verifying, the feature capacity and watermark capacity are measured by the number of bits generated or embedded per pixel (bpp: bit per pixel), and the quality of the watermarked image is measured by its peak signal-to-noise ratio with the original image (in dB ), and select two representative watermark generation and authentication algorithms of [1] and [2] for digital image tampering content recovery:

Literature [1] Z. Qian, G. Feng, X. Zhang, S. Wang. Image self-embedding with high-quality restoration capability, 2011(21): 278-286

Literature [2] He Hong-jie, Zhang Jia-shu, Chen Fen. A self-recovery fragile watermarking scheme for image authentication with superior localization, Science in China Series F-Information Sciences, 2008.51(10): 1487-1507.

1. Feature capacity and reconstruction quality

Feature extraction is an important step in generating restored watermarks. The ideal is to hold as many image features as possible with as little capacity as possible. Using a similar feature extraction method, the more image block features are extracted, the better the image quality will be reconstructed according to the features, but the feature information will also increase. Table 1 presents the statistical results of the feature capacity and reconstruction quality of the present invention and the existing literature [1] and literature [2] for restoring watermarking algorithms. It can be seen from Table 1 that the present invention has the most feature information, and correspondingly, the image quality reconstructed by using features is also the best. Literature [1] has the least feature information, and its reconstruction quality is better, especially for relatively smooth digital images. This is because literature [1] uses the important DCT coefficients of 8×8 image blocks to generate the restored watermark, and the DCT transform It can effectively remove the redundancy between pixels, but due to the large image block, the tampering positioning accuracy of the algorithm is not high, and the quality of the restored image may be reduced. It can also be seen from Table 1 that the feature information extracted by document [1] for different digital images is 1bpp, and the feature information extracted by document [2] for different digital images is 1.5bpp, which shows that documents [1] and [2] ] The extracted feature capacity is fixed. On the contrary, the feature capacity extracted from different digital images by the present invention is variable, and the variation range is 1.80-2.62bpp. The smoother the image, the smaller the extracted feature capacity. the

Table 1 The statistical results of the feature capacity and reconstruction quality of the present invention and existing literature methods

2. Watermark capacity and watermarked image quality

Due to the large watermark capacity of the recoverable watermarking algorithm, in order to minimize the image deformation caused by watermark embedding, the recoverable watermarks are mostly embedded in the low significant bits of the spatial domain. Therefore, the larger the watermark embedding capacity of the recoverable watermarking algorithm, the more serious the image deformation will be, that is, the poorer the quality of the generated watermarked image. Table 2 shows the statistical results of the watermark capacity and watermarked image quality of the present invention and the existing literature [1] and literature [2]. Seen from Table 2, the watermark capacity of documents [1] and [2] are fixed, respectively 3bpp and 1.5bpp, but the watermark capacity of the present invention is variable. The quality of watermarked digital images generated by different images in literature [1] is similar, and their peak signal-to-noise ratios are about 38dB. Literature [2] also has similar quality of watermarked digital images generated by different images, and their peak signal-to-noise ratio are about 48dB respectively. The quality of the watermarked digital images generated by the present invention is quite different from different images, the peak signal-to-noise ratio is between 38-43dB, and the smoother the image, the better the quality of the generated watermarked images. the

Comparing Table 1 and Table 2, it can be seen that the present invention and literature [2] do not add redundant information when embedding watermarks, while literature [1] uses 3 times the capacity of embedded watermarks to locate tampering and improve the quality of tampering recovery. times, the quality of the watermarked image is reduced, and the peak signal-to-noise ratio of the watermarked image generated by the present invention is higher than that of the literature [1]. the

Table 2 Statistical results of watermark capacity and watermarked image quality of the present invention and existing documents

3. General tampering

Fig. 3 shows the tamper detection and tamper recovery results of the present invention and the existing literature [1] and literature [2] under general tampering conditions. Among Fig. 3 (a) part picture is the Beach original gray scale image that size is 328 * 328 pixels, among Fig. 3 (b) part picture is the watermarked digital image that utilizes the algorithm of the present invention to generate, and its peak signal-to-noise ratio is 42.93dB , sub-picture (c) in Figure 3 is a tampered image, which contains the following tampering: add the text "ITP Southwest Jiaotong University"; add a deer below, and the ratio of tampered pixels is about 7.78%. In Fig. 3, (d) sub-graph, (e) sub-graph and (f) sub-graph are respectively the tampering detection results of the two recovery watermarking algorithms of the present invention, existing literature [1] and literature [2]. It can be seen from the figure It is found that the positioning accuracy of literature [1] is lower than that of the present invention and literature [2], because the image block size of the present invention and literature [2] is 2×2 pixels, while the image block size of literature [1] is 8 ×8 pixels. In order to quantitatively compare the tamper detection performance of the algorithm, the missed detection rates of the present invention, the literature [1] and the literature [2] calculated in units of 2×2 pixels are 0.3%, 0% and 3.45%, respectively, and the false detection rates are respectively 0.39%, 5.13% and 0.02%. In Fig. 3, (g) sub-image, (h) sub-image and (i) sub-image are the tampered recovery images of the present invention, document [1] and document [2] respectively, and their peak signal-to-noise ratios with the original image are respectively 43.63dB, 43.06dB and 29.94dB. The recovery effect of the present invention is similar to that of the document [1], both of which have obtained better tampering recovery effects. The recovery effect of literature [2] is poor. There are noise points in the tampered area of (i) in Figure 3. This is because the restored watermark information of these tampered blocks is also destroyed, and literature [2] cannot effectively restore the watermark information caused by altered tampered blocks. the

From the above comparison, it can be seen that under general tampering conditions, the present invention has similar positioning accuracy to document [2], which is better than document [1]. At the same time, the present invention has similar tamper recovery quality to document [1], which is better than Literature [2]. That is, the present invention has good tampering localization accuracy and tampering recovery quality at the same time. the

4. Constant mean value attack

Fig. 4 shows the tampering detection and tampering recovery results of the present invention, the existing document [1] and the document [2] under the condition of mean value attack. Among Fig. 4, (a) sub-picture is the Lena original grayscale image with a size of 256 × 256 pixels, and among Fig. 4 (b) sub-picture is a watermarked digital image generated by the algorithm of the present invention, and its peak signal-to-noise ratio is 42.54dB . Part (c) of Figure 4 is a tampered image, in which a square area with a size of 140×140 pixels is attacked by the mean value, and the tampering ratio is about 30%. Figure 4 (d) sub-graph, (e) sub-graph and (f) sub-graph are the tampering detection results of the present invention, document [1] and document [2] respectively. In Figure 4 (f) there is no white area, indicating that the document [2] cannot detect the tampering, that is, the missed detection rate is 100%, which indicates that the document [2] cannot resist the constant mean attack. Since no tampered block is detected, the algorithm will not perform tampering recovery on the tampered image. The peak signal-to-noise ratio with the original image is 11.20dB. Both the present invention and the document [1] can detect the tampering, and the missed detection rates of the present invention and the document [1] calculated in units of 2×2 pixels are 0.21% and 0%, respectively, and the false detection rates are 0.94% and 0%, respectively. 3.6%. Comparing the sub-graphs (d) and (e) in Figure 4, it can also be seen that there are a few missed image blocks in the tampering detection of the present invention. The attack is carried out on 2×2 image blocks. Document [1] judges tampered blocks in units of 8×8 pixels, so there is no missing image block in the tampered area of Document [1]. In the algorithm tampering detection, the image blocks that are not judged as tampering do not perform the tampering recovery operation, resulting in the tampering recovery quality of the algorithm of the present invention being slightly lower than that of the literature [1]. Parts (g) and (h) in Fig. 4 are tampered restored images of the present invention and literature [1] respectively, and their peak signal-to-noise ratios compared with the original image are 30.81dB and 31.39dB, respectively. It can be seen from the (g) sub-graph in Figure 4 that there are a few noise points (displaying the "hat" area) in the tampered restoration image of the present invention, which is because there are missed detection blocks in this area, and the tampering restoration operation is not performed. of. the

It can be seen from the above comparison that the literature [2] cannot resist the mean value attack on 2×2 blocks. The present invention and document [1] can detect such tampering and effectively restore the tampered area. The positioning accuracy of the present invention is higher than that of document [1], and the false detection rate is lower than that of document [1], but the missed detection rate is slightly higher than that of document [1]. [1], resulting in slightly lower tamper recovery quality than in [1]. the

5. Collage attack

Fig. 5 shows the tamper detection and tamper recovery results of the present invention, the existing document [1] and the document [2] under the collage attack condition. Part (a) and part (b) in Figure 5 are the watermarked Monlisa and Napoleon grayscale images generated with the same key, respectively, and their size is 208×328 pixels. Replace the head area of the watermarked Monelisa image with the same area of the watermarked Napoleon image (collage attack), and the obtained tampered image is shown in Figure 4 (c), and the tampered ratio is about 18%. In Fig. 5, (d), (e) and (f) are the tampering detection results of the present invention, literature [1] and literature [2], respectively, and are calculated in units of 2×2 pixels. The missed detection rates of invention, literature [1] and literature [2] were 2.03%, 86.67% and 20.22%, respectively, and the false detection rates were 0.13%, 2.27% and 0.26%. It can be seen from the part (e) in Figure 5 that literature [1] only detects the boundary of the collage area, and the interior of the collage area is certified. The reason why the literature [1] can detect the 8×8 image blocks located at the boundary of the collage area is that the collage attack only changes some pixels in these image blocks. Since the image blocks that have not been detected for tampering do not perform tampering recovery operations, the tampering recovery quality of the literature [1] is low, as shown in the sub-graph (h) of Figure 5, and its peak signal-to-noise ratio is 17.72dB. The present invention and literature [2] can resist the collage attack. Comparing the (d) sub-graph and (f) sub-graph in Figure 5, the ratio of the black points in the collage area in (d) sub-graph to (f) in the sub-graph Many, which shows that the missed detection rate of the present invention is lower than that of literature [2]. The invention can effectively restore the image blocks whose watermark information has been tampered with, so the quality of the tampering recovery of the invention is obviously better than that of literature [2]. Part (g) and part (i) in Fig. 5 are tampered recovery images of the present invention and literature [2] respectively, and their peak signal-to-noise ratios are 33.82dB and 24.83dB respectively. the

It can be seen from the above comparison that the document [1] cannot resist the collage attack, but the present invention and the document [2] can resist the collage attack. the

Table 3 shows the tampering detection performance (false detection rate) and tampering restoration quality (tampering restoration image and watermarked image The comparison results of the peak signal-to-noise ratio). In summary, it can be seen that the restored watermark capacity in the algorithm of the present invention is variable. Different volumes of restored watermarks are generated for different digital images. The smoother the image, the smaller the amount of restored watermarks generated. Under the premise of satisfying the preservation of the image block information, the watermark embedding capacity is reduced as much as possible, thereby improving the quality of the watermarked image. At the same time, the algorithm of the invention has higher tampering location performance and tampering recovery quality. Under general tampering, mean attack, and collage attack, it can accurately locate the tampered position, detect tampered blocks with high probability, and restore tampered images with high quality. the

Table 3 Comparison of tamper detection performance and tamper recovery quality

Note: False detection rate = tampering ratio × missed detection rate + (1- tampering ratio) × false detection rate.

Claims (1)

1. A variable-capacity watermark generation and authentication method with recoverable tampering content of a digital image, comprising the following steps: A. Watermark generation and embedding A1. Image segmentation and classification: Divide the original image X with a size of 2m×2n into m×n non-overlapping 2×2 image blocks X _i ={ _xi1 , _xi2 , _xi3 , _xi4 }, Wherein, i is the image block number, i=1, 2,..., N, N=m×n is the number of image blocks; and the image blocks are divided into smooth image blocks and non-smooth image blocks according to the content of the image blocks kind; A2. Pseudo-random sequence and block chain generation: Generate a real-valued pseudo-random sequence of length N based on the user key Key R={r _i |i=1, 2,...,N}, ordered by the index of R Sequence generation block chain {(X _i ,X _i′ )|i, i′∈[1,N]}, i′ is the value of the i-th element in the R index ordered sequence, and X _i’ is the mapping of X _i piece; A3. Variable capacity feature extraction: calculate and generate v-bit block features F _i ={f _i1 , f _i2 ,..., f _iv } for each image block X _i , where f _i2 ~ f _i6 are image blocks X _i The 5-bit binary code of the upper 5-bit average value, if the image block X _i is a smooth image block, then v=6 and f _i1 =0, otherwise v=12, f _i1 =1, f _i7 ~ f _i12 are non-smooth images the detail encoding of the block; A4. Variable-capacity watermark generation: use the i-th random number r _i of the pseudo-random sequence R to generate a binary pseudo-random sequence B _i ={b _ij |j=1, 2,...12}, and encrypt image block features F _i = {f _i1 , f _i2 , ..., f _iv } generate restored watermark W _i = {w _i1 , w _i2 , ..., w _iv },

in, is an XOR operation; the value range of subscript j for smooth image block b _ij is 1-6, and the value range of subscript j for non-smooth image block b _ij is 1-12; A5 _. Variable-capacity watermark embedding: Embed the restored watermark W _i of the image block Xi into the least significant bit of its mapping block Xi _' to generate a watermarked image block Y _i' , If _Xi is a non-smooth image block, If _Xi is a smooth image patch,

B. Watermark extraction and comparison B1. Blocking, pseudo-random sequence and block chain generation: Y ^* is the tested image received after the watermarked image Y is transmitted, and Y ^* is divided into 2×2 blocks according to step A1

Then follow the same operation and key Key in step A2 to generate random sequence R and block chain

B2. Block feature calculation and reconstruction: for each tested image block

Calculate the image block features according to the A3 step operation

while mapping image blocks according to Recovered Watermark Extracted by Least Significant Bits

Refactor block feature

Among them, B _i is a binary pseudo-random sequence of length 12 generated based on r _i , from which the image block is mapped

Low bit extraction watermark information

According to the following formula, B3. Classification feature comparison: compare the tested image blocks

Block characteristics obtained by high-level calculation

and map blocks from it Extraction of Least Significant Bits to Recover Block Features of Watermark Reconstruction

Generate comparison matrix D={d _i |i=1, 2,..., N} according to the following formula, in

C. Tamper Detection C1. Calculate and generate the neighborhood characteristic matrix Δ={δ _i |i=1, 2, ..., N} of the comparison matrix D, δ _i ＝∑d _j , j=i±1, i±n, i+n±1, in±1 C2. According to the comparison matrix D and the neighborhood feature matrix Δ, generate the initial state tampering detection matrix T ⁰

C3. Generate a tampering detection matrix T according to the initial state tampering detection matrix ^T0 generated in the C2 step,

in j=i±1, i±n, i±n±1, t _i =1 means the image block under test

tampered, t _i =0 means the image block under test

It is true; D. Tamper recovery If the elements in T are not all 0, it indicates that there is a tampered image block in the tested image, then perform the following two steps in turn on the tested image to obtain the tampered recovery image Y ^R ; D1. Feature recovery: For image blocks judged to be tampered If its mapped block Y _i' ^* is real, reconstruct block features with B2 step

right

to restore,

Among them, Γ ^-1 () is the inverse function of the feature encoding function of the image block; at the same time, a label matrix L is generated to mark the tampered block that has not been restored,

D2. Neighborhood recovery: if the elements in L are not all 0, the image block corresponding to l _i =1

Correct the image block by using the mean value of the effective pixels in the 12 adjacent pixels The value of each pixel in .

Images