WO2020051953A1

WO2020051953A1 - Reversible image watermarking method and apparatus based on prediction additive error expansion

Info

Publication number: WO2020051953A1
Application number: PCT/CN2018/107911
Authority: WO
Inventors: 翟懿奎; 邓文博; 徐颖; 王天雷; 曹鹤; 甘俊英; 曾军英; 秦传波
Original assignee: 五邑大学
Priority date: 2018-09-10
Filing date: 2018-09-27
Publication date: 2020-03-19
Also published as: CN109360140A; CN109360140B

Abstract

Provided are a reversible image watermarking method and apparatus based on prediction additive error expansion; first, a predictor is created according to a pixel x[i, j] and known pixels around same; then, the predictor is used to calculate the prediction error of the pixel x[i, j]; finally, an additive expansion algorithm is used to expand the prediction error of a specific value; therefore, the objective of translational motion is achieved by means of performing addition on the prediction error, thus space is made for expansion and watermark data is embedded; a relatively small space is taken up, and the problem of pixel value overflow is avoided, thereby significantly reducing the visual distortion caused by the watermark.

Description

Reversible image watermarking method and device based on prediction error addition extension

Technical field

The invention relates to the technical field of reversible image watermarking, in particular to a reversible image watermarking method and device based on prediction error addition extension.

Background technique

Reversible watermarking is the current research hotspot of digital watermarking technology. Compared with traditional digital watermarking technology, reversible watermarking can completely restore the original host information without distortion, which has great research value and good application prospects, especially for original host information. Application areas with high fidelity requirements, such as aerial photography information collection. Most of the existing reversible watermarking technologies are based on displacement expansion to embed watermark data. When this method has the problem of overflowing pixel values, it is easy to cause a significant reduction in the visual quality of the image.

Summary of the Invention

In order to overcome the shortcomings of the prior art, an object of the present invention is to provide a reversible image watermarking method and device based on prediction error addition expansion. Watermarking data is embedded by using addition operation instead of traditional displacement operation to avoid pixel value overflow. Problem, which can significantly reduce the visual distortion caused by the watermark.

The technical solutions adopted by the present invention to solve its problems are:

A reversible image watermarking method based on the prediction error addition extension includes the following steps:

S1. Scan the image to obtain the current pixel x [i, j] and known pixels around it;

S2. Establish a predictor based on the current pixel x [i, j] and known pixels around it;

S3. Use a predictor to calculate the prediction error of the current pixel x [i, j];

S4. The prediction error of the specific value is extended by the addition extension algorithm to complete the watermark embedding process of the image.

Further, in step S2, a predictor is established based on the current pixel x [i, j] and known pixels around it, including the following steps:

S21. The current pixel x [i, j] and the surrounding known pixels form the context of the predictor;

S22. The expression for establishing the predictor according to the context of the predictor is:

among them,

Is the predicted value of pixel x [i, j], and H and W are the height and width of the image, respectively.

Further, the prediction error in step S3 is obtained by the following formula:

Among them, e [i, j] is the prediction error of pixel x [i, j].

Further, the formula of the addition expansion algorithm in step S4 is:

Among them, sign (e) is a sign function, b is binary data to be embedded, and both ME and LE are pre-set values.

Further, the formula for the sign function sign (e) is:

Further, the formula for extracting the inverse operation of the addition extension algorithm is:

A device for storing a reversible image watermarking method based on prediction error addition and expansion includes a control module and a storage medium for storing control instructions. The control module reads the control instructions in the storage medium and executes the following steps:

Q1. Scan the image to get the current pixel x [i, j] and known pixels around it;

Q2. Establish a predictor based on the current pixel x [i, j] and known pixels around it;

Q3. Use the predictor to calculate the prediction error of the current pixel x [i, j];

Q4. Use the addition and expansion algorithm to extend the prediction error of a specific value to complete the watermark embedding process of the image.

Further, when the control module executes step Q2, the predictor is established based on the current pixel x [i, j] and known pixels around it, including the following steps:

Q21: The current pixel x [i, j] and the surrounding known pixels form the context of the predictor;

Q22. The expression for establishing a predictor based on the context of the predictor is:

among them,

Further, when the control module executes step Q3, it uses a predictor to calculate the prediction error of the current pixel x [i, j], and the prediction error is obtained by the following formula:

Among them, e [i, j] is the prediction error of pixel x [i, j].

Further, when the control module executes step Q4, the prediction error of the specific value is extended by the addition extension algorithm, and the formula of the addition extension algorithm is:

The beneficial effects of the present invention are: a reversible image watermarking method and device based on prediction error addition expansion. In the traditional displacement expansion method, a position table is usually used to distinguish the expanded difference values. The difference or the difference of each pixel requires one bit to record, so the overhead of the position table is often large. Although the position table is highly compressible, even the compressed position table will still occupy a large amount. space. The present invention first establishes a predictor based on a pixel x [i, j] and surrounding known pixels, then uses the predictor to calculate the prediction error of the pixel x [i, j], and finally uses an addition expansion algorithm to predict a specific value The error is expanded. Therefore, by adding the prediction error to the translation to achieve the purpose of translation, the watermark data can be embedded for space expansion, which not only takes up a relatively small space, but also avoids the problem of pixel value overflow. , Which can significantly reduce the visual distortion caused by the watermark.

BRIEF DESCRIPTION OF THE DRAWINGS

The invention is further described below with reference to the drawings and examples.

FIG. 1 is a flowchart of a prediction error addition and extension method of the present invention;

FIG. 2 is a context diagram of a predictor;

FIG. 3 is a schematic diagram of a prediction error addition and extension method of the present invention.

detailed description

Referring to FIGS. 1-3, a reversible image watermarking method based on prediction error addition and extension includes the following steps:

In step S2, a predictor is established based on the current pixel x [i, j] and known pixels around it, including the following steps:

among them,

Is the predicted value of pixel x [i, j], H and W are the height and width of the image, respectively.

The prediction error in step S3 is obtained by the following formula:

Among them, e [i, j] is the prediction error of pixel x [i, j].

The formula of the addition and expansion algorithm in step S4 is:

Among them, sign (e) is the sign function, b is the binary data to be embedded, and both ME and LE are pre-set values.

Specifically, the formula of the sign function sign (e) is:

Among them, the formula of the extraction inverse operation of the addition extension algorithm is:

In addition, a device for storing a reversible image watermarking method based on prediction error addition expansion includes a control module and a storage medium for storing a control instruction. The control module reads the control instruction in the storage medium and executes the following steps:

Wherein, when the control module executes step Q2, the predictor is established based on the current pixel x [i, j] and known pixels around it, including the following steps:

among them,

Wherein, when the control module executes step Q3, it uses a predictor to calculate the prediction error of the current pixel x [i, j], and the prediction error is obtained by the following formula:

Among them, e [i, j] is the prediction error of pixel x [i, j].

Wherein, when the control module executes step Q4, the prediction error of the specific value is extended by the addition extension algorithm, and the formula of the addition extension algorithm is:

Among them, the formula of the sign function sign (e) is:

The storage medium also stores an extraction inverse operation algorithm corresponding to the addition expansion algorithm, and its formula is:

Specifically, in the traditional displacement expansion method, a position table is usually used to distinguish the expanded difference value. Since each difference value or the difference value of each pixel in the position table requires one bit to record, Therefore, the cost of the position table is often large. Although the position table is highly compressible, even the compressed position table still takes up a relatively large amount of space. The present invention first establishes a predictor based on a pixel x [i, j] and surrounding known pixels, then uses the predictor to calculate the prediction error of the pixel x [i, j], and finally uses an addition expansion algorithm to predict a specific value The error is expanded. Therefore, by adding the prediction error to the translation to achieve the purpose of translation, the watermark data can be embedded for space expansion, which not only takes up a relatively small space, but also avoids the problem of pixel value overflow. , Which can significantly reduce the visual distortion caused by the watermark.

Specifically, the traditional difference expansion takes the difference between pixels as the expansion object and uses a shift operation to expand. The digital form can be expressed as d ′ = 2 × d + b, where b is the binary data to be embedded, such as 0 or 1, and d is the difference between the two pixels d = p ₁ -p ₂ . In the prediction error addition and extension method of the present invention, the prediction error is

among them,

Is the predicted value of pixel x [i, j], and the predicted value is the guess value of the current pixel x [i, j] calculated by the predictor according to the context. For the context of a simple predictor, it consists of the current pixel and known pixels around it, as shown in Figure 2. Each square in Figure 2 represents a pixel, the square corresponding to x [i, j] is the pixel to be predicted, and the four pixels above and to the left are the known pixels that make up the context of the predictor, so The mathematical expression of the predictor is:

Predictors usually only use pixels that have been processed previously. In this way, when extracting a watermark, as long as the image is processed using the same scanning order as when the watermark is embedded, the same predictor context can be reconstructed and embedded. Restore its predicted value to meet the reversible requirements of the watermark method. The predictor can use the current pixel and all pixels before it, not just a specific pixel, and a well-designed predictor is obtained through a large number of experiments and linear regression, so it can achieve a relatively high prediction accuracy. Therefore, the prediction error is generally smaller than the difference between pixels, so the prediction error is more suitable for expansion processing. Therefore, it is feasible and advantageous to use the prediction error instead of the difference between pixels for expansion.

The additive expansion algorithm of the present invention makes room for expansion by translating the prediction error, and each expansion operation is performed only on the prediction error of a specific value. For example, if the absolute value of the prediction error to be expanded is set to 1, then the prediction error with an absolute value greater than 2 will be increased by 1 when performing a translation, so that there will be no prediction error with an absolute value of 2. Then, if the bit to be embedded is 0, the prediction error to be extended is kept unchanged, that is, 1; if the bit to be embedded is 1, the prediction error to be extended is increased by 1, so that it becomes 2. When the watermark is extracted, if the absolute value of the prediction error is greater than 2, it means that it is not embedded in the watermark, and it can be restored to the original prediction error by subtracting 1; if the absolute value of the prediction error is 2, it means that one is embedded as 1. If the absolute value of the prediction error is 1, it means that a bit of 0 is embedded. Since 0 is extracted, the prediction error is Nothing has changed, so no action is required on it. In this way, as long as the value of 1 is recorded, the prediction error between embedded data and non-embedded data can be distinguished without using a data structure such as a position table. In the addition expansion algorithm, because the distribution of the prediction error is highly concentrated in a region with a very small absolute value, even if it is only a prediction error equal to a specific value, such as a prediction error with an absolute value of 1, the number can still account for a relatively large part of the total .

Therefore, based on the above principles, the formula of the addition expansion algorithm can be obtained as:

In addition, the formula for the extraction inverse operation of the addition extension algorithm is:

Where sign (e) is a sign function, and

b is the binary data to be embedded; both ME and LE are predetermined values, and the selection rules of ME and LE are: maximize the number of prediction errors that satisfy | e | = ME, and make predictions that satisfy | e | = LE The number of errors is the smallest, usually zero. In general, ME is a small value, such as ME = 1; while LE is a large value, that is, LE> ME. If you encounter a special situation, that is, LE <ME, you only need to adjust the addition expansion algorithm to the following formula:

Since ME and LE must be determined before watermark extraction, they can be transmitted as a key with the watermarked image, or they can be embedded in the image using other watermark embedding methods, which can effectively prevent others from stealing information. If the difference between the original image and the predicted image is regarded as an error image, and the distribution of the prediction error is regarded as a histogram of the error image, then the additive expansion method based on the prediction error is similar to the watermark based on the histogram operation The embedding method is shown in Figure 3. In Figure 3, ME corresponds to the highest point of the histogram, while LE corresponds to the lowest point of the histogram. The addition expansion method is to first shift the histogram column between ME and LE to the right by 1 and then use the two histogram columns corresponding to ME and ME + 1 to embed 0 or 1 respectively.

In addition, the problem of pixel overflow also needs to be considered when performing the expansion processing. For example, a pixel with a value of 0 cannot be reduced by 1, and a pixel with a value of 255 cannot be increased by 1. Therefore, the addition and expansion method in the present invention performs watermark embedding only on pixels with a value between 1 and 254. Similarly, when performing watermark extraction, only pixels with a value between 1 and 254 are performed. If pixels with an original value of 1 or 254 are extended by addition to 0 and 255, they will be recorded at the head of the watermark information. Therefore, when a pixel with a value of 0 or 255 is encountered, it is first determined whether it is changed through 1 and 254, and then how to operate it. Since the addition and expansion method in the present invention only increases the pixel by one at most, and the edge pixels with values of 255 and 0 occupy a very limited proportion in the image of the real application scene, this overhead is very small. Therefore, the present invention The method of addition and expansion not only can provide larger embedding capacity, but also can avoid the problem of pixel value overflow.

In addition, the fidelity of the image processed by the addition expansion method in the present invention is also higher than that of the image using the conventional displacement-based difference expansion method. It can be known from the mathematical expression of displacement expansion that the changes brought by it are:

The addition expansion method in the present invention is:

So the changes it brings are just:

Since b takes 0 or 1, the addition expansion method in the present invention will only increase the pixel value by at most 1, so compared with the traditional displacement-based difference expansion method, the distortion caused by the addition expansion method in the present invention smaller.

The above is a detailed description of the preferred implementation of the present invention, but the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or replacements without departing from the spirit of the present invention. Equivalent deformations or substitutions are included in the scope defined by the claims of the present application.

Claims

A reversible image watermarking method based on the prediction error addition extension is characterized in that it includes the following steps:

S1. Scan the image to obtain the current pixel x [i, j] and known pixels around it;

S2. Establish a predictor based on the current pixel x [i ,, j] and known pixels around it;

S3. Use the predictor to calculate the prediction error of the current pixel x [i ,, j];

S4. The prediction error of the specific value is extended by the addition extension algorithm to complete the watermark embedding process of the image.
The reversible image watermarking method based on prediction error addition expansion according to claim 1, characterized in that: in step S2, a predictor is established based on the current pixel x [i, j] and known pixels around it, comprising the following steps: :

S21. The current pixel x [i, j] and the surrounding known pixels form the context of the predictor;

S22. The expression for establishing the predictor according to the context of the predictor is:

2≤i≤H, 2≤j≤W-1

among them,
Is the predicted value of pixel x [i, j], and H and W are the height and width of the image, respectively.
The reversible image watermarking method based on the prediction error addition expansion according to claim 2, wherein the prediction error in the step S3 is obtained by the following formula:

Among them, e [i, j] is the prediction error of pixel x [i, j].
The reversible image watermarking method based on prediction error addition expansion according to claim 3, wherein the formula of the addition expansion algorithm in step S4 is:

Among them, sign (e) is a sign function, b is binary data to be embedded, and both ME and LE are pre-set values.
The reversible image watermarking method based on the prediction error addition expansion according to claim 4, wherein the formula of the sign function sign (e) is:
The reversible image watermarking method based on prediction error addition expansion according to claim 5, wherein the formula for extracting the inverse operation of the addition extension algorithm is:
A device for storing a reversible image watermarking method based on prediction error addition expansion, characterized in that it includes a control module and a storage medium for storing control instructions, the control module reads the control instructions in the storage medium and executes the following step:

Q1. Scan the image to get the current pixel x [i, j] and known pixels around it;

Q2. Establish a predictor based on the current pixel x [i, j] and known pixels around it;

Q3. Use the predictor to calculate the prediction error of the current pixel x [i ,, j];

Q4. Use the addition and expansion algorithm to extend the prediction error of a specific value to complete the watermark embedding process of the image.
The device according to claim 7, characterized in that, when the control module executes step Q2, establishing a predictor based on the current pixel x [i, j] and known pixels around it, comprising the following steps:

Q21: The current pixel x [i, j] and the surrounding known pixels form the context of the predictor;

Q22. The expression for establishing a predictor based on the context of the predictor is:

2≤i≤H, 2≤j≤W-1

among them,
Is the predicted value of pixel x [i, j], and H and W are the height and width of the image, respectively.
The device according to claim 8, wherein when the control module executes step Q3, a predictor is used to calculate a prediction error of the current pixel x [i, j], and the prediction error is obtained by the following formula:

Among them, e [i, j] is the prediction error of pixel x [i, j].
The device according to claim 9, characterized in that, when the control module executes step Q4, the prediction error of a specific value is extended by an addition extension algorithm, and the formula of the addition extension algorithm is:

Among them, sign (e) is a sign function, b is binary data to be embedded, and both ME and LE are pre-set values.