CN115131189A - Image reversible information hiding method and system based on convolutional neural network - Google Patents

Abstract

The present application provides an image reversible information hiding method and system based on a convolutional neural network, and relates to the technical field of reversible information hiding. When performing reversible information hiding, the method divides an original image into mutually independent cross-set images and point-set images, The cross-set image and the point-set image are predicted respectively, and the reversible information-hiding image is generated based on the difference information between the original image and the predicted image. And in the prediction process, the image features of different size receptive fields of the image are extracted, and the channels are superimposed, which can make more full use of the correlation between pixels, improve the prediction accuracy of the target pixel, and greatly increase the embedding capacity.

Description

A method and system for image reversible information hiding based on convolutional neural network

technical field

The present application relates to the technical field of reversible information hiding, and in particular, to a method and system for image reversible information hiding based on convolutional neural networks.

Background technique

The statements in this section merely provide background information related to the present application and do not necessarily constitute prior art.

Reversible information hiding can embed information into the original image, and restore the original image and information losslessly after decrypting the information and extracting the information. Based on these two advantages, reversible information hiding is widely used in military, medical and super-resolution processing fields.

At present, there are many methods for reversible information hiding (RDH). One type is mainly to improve the embedding method, such as the method of difference expansion, histogram translation and error prediction expansion; the other type is mainly to improve the prediction method, and improve the accuracy of prediction by continuously improving the performance of prediction, such as Difference predictor, median edge direction predictor, gradient adaptive predictor, bilinear interpolation predictor, etc.

The inventor found that in the process of reversible information hiding, the traditional pixel prediction method adopts linearity and only uses one or several adjacent pixels to predict the image target pixel, which cannot use the correlation of more adjacent pixels, resulting in the accuracy of prediction. is not high, so there is more reversible information hiding space.

SUMMARY OF THE INVENTION

Aiming at the problems of low target pixel prediction accuracy and low embedding capacity in the existing methods, the present application provides an image reversible information hiding method and system based on a convolutional neural network.

Specifically, the following technical solutions are adopted:

In a first aspect, an embodiment of the present application provides an image reversible information hiding method based on a convolutional neural network, including:

Divide the original image into independent cross-set images and point-set images;

Use the point set image predictor to extract the image features of the receptive fields of different sizes of the cross set image, and perform channel stacking to obtain the predicted point set image; determine the first difference between the point set image and the predicted point set image information, embed the secret information into the first difference information, and add it with the predicted point set image to obtain the first secret-carrying image;

Use the fork set image predictor to extract the image features of the receptive fields of different sizes of the first secret image, and perform channel stacking to obtain the predicted fork set image; determine the second difference between the fork set image and the predicted fork set image difference information, embedding secret information into the second difference information, and adding the predicted cross-set image to obtain a second secret-carrying image;

A reversible information hiding image is obtained by fusing the first secret-carrying image and the second secret-carrying image.

In a possible implementation, the point set image predictor and the cross set image predictor have the same structure, and both use a convolutional neural network; the convolution kernels of different sizes of the convolutional neural network are used to extract the Image features, the receptive field is expanded according to the superimposed combination of image features, and the target image includes a cross-set image and a point-set image.

In a possible implementation, the convolutional neural network includes a backbone prediction network and an auxiliary prediction network. In the backbone prediction network, the target image outputs images with different sizes of receptive fields through the convolutional network of the Inception structure feature, the image features are superimposed on the channel dimension, and the image features on different channels are used to predict the target pixels to obtain the main prediction image; in the auxiliary prediction network, the target image is predicted by a convolution kernel of a specific size, and the target pixel is reduced. Dimension to obtain an auxiliary predicted image; the predicted image is determined according to the main predicted image and the auxiliary predicted image.

In a possible implementation manner, a channel attention mechanism is added after the backbone prediction network superimposes the channel dimension of the image features, so as to adjust the weights of the image features on different channels.

In a possible implementation, the channel attention mechanism includes ECA-Net; after ECA-Net aggregates convolution features using GAP without dimensionality reduction, the size of one-dimensional convolution kernel is adaptively selected to learn channel attention .

In a possible implementation, the secret information is embedded in the difference information in the following manner, the difference information including the first interpolation information and the second difference information:

The target pixel ui _{,j is predicted by the pixel v i,j in} the area, and the predicted pixel can be expressed as: u' _i,j =CNN(vi _,j );

Obtain d i,j by making the difference between the predicted pixels u' _i,j and u _{i ,j} : d _i,j =u' _i,j -u _i,j ;

Information embedding process: D _i,j =2d _i,j +b;

Among them, D _i,j is the prediction error after expansion, b is the payload to be embedded, and then D _i,j is changed according to the histogram translation method; after the information is embedded, u _i,j is calculated as U _i,j : U _i,j =D _i,j +u'_i,j;

Using the predicted value u' _i,j and the modified pixel value U _i,j , the modified prediction error value is calculated by the following formula: D _i,j =U _i,j -u'_i,j;

The value of the secret information is: b=D _i,j mod2;

The original pixel value is:

In a possible implementation manner, information extraction is performed on the reversible information hiding image in the following manner:

Divide the reversible information hiding image into independent secret-carrying point set images and secret-carrying fork-set images; obtain the predicted secret-carrying fork-set images according to the secret-carrying point set images and fork-set image predictors; determine the secret-carrying fork set images and a second carrier difference image between the predicted fork set images, and secret information and fork set images are extracted from the second carrier difference image;

According to the cross set image and the point set image predictor, obtain the predicted secret point set image; determine the first secret point set image and the predicted secret point set image between the first secret point set image Extract secret information and point set image from secret difference image;

The original image is obtained by fusing the cross set image and the point set image.

In a second aspect, an embodiment of the present application provides an image reversible information hiding system based on a convolutional neural network, including:

The image division module is used to divide the original image into mutually independent cross-set images and point-set images;

The first secret image acquisition module is used to extract the image features of the receptive fields of different sizes of the fork set image by using the point set image predictor, and perform channel stacking to obtain the predicted point set image; determine the point set image and the predicted image feature. the first difference value information between the point set images, the secret information is embedded in the first difference value information, and added with the predicted point set image to obtain the first secret-carrying image;

The second secret image acquisition module uses a fork set image predictor to extract the image features of the receptive fields of different sizes of the first secret image, and performs channel stacking to obtain a predicted fork set image; the second difference information between the cross-set images, embedding secret information into the second difference information, and adding the predicted cross-set image to obtain a second secret-carrying image;

The fusion module is used for fusing the first secret-carrying image and the second secret-carrying image to obtain a reversible information hiding image.

In a third aspect, an embodiment of the present application provides a computer device, including: a processor, a memory, and a bus, where the memory stores machine-readable instructions executable by the processor, and when the computer device runs, the processor It communicates with the memory through a bus, and when the machine-readable instructions are executed by the processor, execute the convolutional neural network-based convolutional neural network described in the first aspect and any possible implementation manner of the first aspect. Steps of an image reversible information hiding method.

In a fourth aspect, embodiments of the present application provide a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is run by a processor, any of the above-mentioned first aspect and the first aspect is executed. The steps of the convolutional neural network-based image reversible information hiding method described in a possible implementation manner.

The beneficial effects of this application are:

1. When performing reversible information hiding in this application, the original image is divided into a cross-set image and a point-set image that are independent of each other, and the cross-set image and the point-set image are predicted respectively, based on the difference information between the original image and the predicted image. Generate a reversible information hiding image, and in the prediction process, extract the image features of different sizes of receptive fields of the image, and perform channel stacking, which can make more full use of the correlation between pixels and improve the prediction accuracy of the target pixel, which can greatly improve the embedding. capacity.

2. Compared with traditional error prediction predictors such as MED and other predictors, the convolutional deep network error prediction under the same image not only expands the range of the receptive field in space, but also makes higher use of the correlation between pixels. The attention mechanism is added to the channels of the convolutional neural network to enhance the correlation between pixels between channels, improve the prediction performance of the convolutional neural network, and increase the embedding capacity of information.

Description of drawings

The accompanying drawings forming a part of the present application are used to provide further understanding of the present invention, and the exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention.

1 is a schematic flowchart of an image reversible information hiding method based on a convolutional neural network provided by an embodiment of the present invention;

2 is a schematic diagram of an information embedding process provided by an embodiment of the present application;

3 is a schematic diagram of the network structure of a convolutional neural network predictor provided by an embodiment of the present application;

Fig. 4 is the schematic flow chart of the ECA-Net network structure provided by the embodiment of the present application;

5 is a schematic diagram of an information extraction process flow provided by an embodiment of the present application;

6 is a histogram of the first error distribution of a convolutional neural network provided by an embodiment of the present application;

7 is a second error distribution histogram of a convolutional neural network provided by an embodiment of the present application;

Fig. 8 is the traditional MED error distribution histogram provided by the embodiment of the present application;

9 is a schematic structural diagram of an image reversible information hiding system based on a convolutional neural network provided by an embodiment of the present application;

FIG. 10 is a schematic structural diagram of a computer device provided by an embodiment of the present application.

Detailed ways

The present application will be further described below with reference to the accompanying drawings and embodiments.

It should be noted that the following detailed description is exemplary and intended to provide further explanation of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

It should be noted that the terminology used herein is for the purpose of describing specific embodiments only, and is not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly dictates otherwise, the singular is intended to include the plural as well, furthermore, it is to be understood that when the terms "comprising" and/or "including" are used in this specification, it indicates that There are features, steps, operations, devices, components and/or combinations thereof.

Example 1

As shown in Figure 1, based on the problems of low target pixel prediction accuracy and low embedding capacity in traditional predictors, the present application provides a convolutional neural network-based image reversible information hiding method, which specifically includes the following steps:

S101: Divide the original image into mutually independent cross-set images and point-set images.

In a specific implementation, the original image is divided into non-overlapping image blocks of the same size, and the image blocks are divided into mutually independent cross-set images and point-set images, and changes in pixels in one set will not affect the other set.

S102: Use a point set image predictor to extract image features of different sizes of receptive fields of the cross set image, and perform channel stacking to obtain a predicted point set image; determine the first point set image between the point set image and the predicted point set image. difference information, embedding secret information into the first difference information, and adding it with the predicted point set image to obtain a first secret-carrying image;

S103: Use a fork set image predictor to extract image features of different sizes of receptive fields of the first secret-carrying image, and perform channel stacking to obtain a predicted fork set image; determine the difference between the fork set image and the predicted fork set image second difference information, embedding secret information into the second difference information, and adding the predicted cross-set image to obtain a second secret-carrying image;

S104: Fusing the first secret-carrying image and the second secret-carrying image to obtain a reversible information hiding image.

In a specific implementation, as shown in Figure 2, the information embedding process is outlined as follows:

(1) First, the image I is divided into a checkerboard-type "point" set P1 and a "fork" set F1.

(2) Use the convolutional neural network to predict the set of "points" P1 in the first-order "fork" set F1, and make a difference between the predicted set of "points" and the original set of "points" to obtain the difference information M1, and the secret information S1 Embed into the difference information M1, and then add the difference information M1 to the predicted "point" set image to obtain the secret "point" set P1'; the second stage uses the secret "point" set P1' to predict the "fork" set F1, the predicted "fork" set is compared with the original "fork" set to obtain the difference information M2, the secret information S2 is embedded into the difference information M2 using the error prediction extension technique, and then the difference information M2 is added to the predicted "Fork" set image F1'.

(3) The final secret-carrying image Im is obtained after adding the secret information embedded in two stages respectively.

In this way, when performing reversible information hiding, the original image is divided into mutually independent cross-set images and point-set images, and the cross-set images and point-set images are predicted respectively, and a reversible image is generated based on the difference information between the original image and the predicted image. The information hides the image, and in the prediction process, the image features of different sizes of receptive fields of the image are extracted, channel stacking is performed, and the prediction accuracy of the target pixel is improved by nonlinear transformation with a larger range of surrounding pixels, thereby greatly improving the embedding capacity.

As an optional implementation, in order to improve the prediction accuracy of the convolutional neural network for the target pixel, the point set image predictor and the cross-set image predictor have the same structure, and both use a convolutional neural network; Convolution kernels of different sizes extract image features of the target image respectively, and expand the receptive field according to the superimposed combination of image features. The target image includes a cross-set image and a first secret image.

Optionally, the convolutional neural network includes a backbone prediction network and an auxiliary prediction network. In the backbone prediction network, the target image outputs a variety of image features of different sizes and different receptive fields through the convolutional network of the Inception structure. The channel dimension is superimposed, and the image features on different channels are used to predict the target pixels to obtain the main prediction image; in the auxiliary prediction network, the target image is predicted by a convolution kernel of a certain size, and the auxiliary prediction image is obtained by reducing the dimension. ; determining the predicted image according to the main predicted image and the auxiliary predicted image.

In specific implementation, this embodiment uses two predictors, the main prediction and the auxiliary prediction, to predict the target pixel, and takes the point set image to predict the cross set image as an example, which specifically includes the following steps:

(1) The target image is first spliced with a convolution kernel size of 5*5 and 3*3 to obtain a 7*7 receptive field, and the image features are initially extracted.

(2) Backbone prediction network: The extracted image features enter the convolutional network of the Inception structure. After 4 convolution kernels with a size of 1*1 and a channel number of 32, the dimensionality reduction is carried out, and the network structure can be more compact after dimensionality reduction. The image features after dimensionality reduction are combined with convolution kernel sizes of 3*3, 5*5, and 7*7, respectively, and the receptive fields can be obtained with sizes of 5*5, 7*7, and 9*9 respectively. receptive field.

(3) For the image features of the four different receptive fields output by the Inception structure, splicing and stacking in the second dimension (32+32+32+32=128) will obtain 128-dimensional image features, that is, the number of image channels. is 128.

(4) As an optional embodiment, in order to make full use of the image features superimposed by different receptive fields in the channel dimension, a channel attention mechanism is added to adjust the weights of image features on different channels. Optionally, this embodiment introduces ECA-Net to strengthen the weight of feature images on important channels, and similarly, reduces the weight of channels with weak image feature expression.

(5) Use 5*5 and 3*3 convolution to check the target pixel prediction for the important channel feature image extracted by ECA-Net, and obtain the backbone prediction "fork" set image, and calculate the backbone with the target "fork" set image MSELoss1.

(6) Auxiliary prediction network: The input convolution size of the feature image output in (1) is 5*5 convolution pixel prediction, and after 1*1 convolution dimension reduction, the auxiliary prediction "cross" set image is obtained, which is consistent with the target "cross". "Auxiliary MSELoss2 for Set Image Computation.

(7) Finally, the Loss of the predicted "fork" set image and the target "fork" set image is:

MSELoss=MSELoss1+λ×MSELoss2; (λ is a constant, λ=0.2)

By training a convolutional neural network predictor to take advantage of the strong correlation between pixels, half of the pixels predict the other half of the target pixel value. The receptive field ranges of 5*5, 7*7, 9*9, and 11*11 of the convolutional neural network are obtained by using convolution kernel sizes of 3*3, 5*5, and 7*7 for different stacking combinations. The convolutional neural network structure is shown in Figure 3.

In order to improve the performance of the convolutional neural network, the ECA-Net channel attention network structure used in this embodiment is shown in Figure 4. The channel attention network is used to calculate a weight, and the weight is calculated with the feature map. The channel features of the feature map are corrected for changes.

Here, ECA-Net uses GAP without dimensionality reduction to aggregate convolution features, firstly uses an adaptive convolution kernel of size K, then performs one-dimensional convolution, and then performs Sigmoid learning channel attention. For efficiency and effectiveness, channel attention is learned using the frequency band matrix W ^k using local cross-channel interactions. The specific representation of ^Wk is as follows:

Among them, K is the size of the convolution kernel, and C is the number of channels.

output=σ(C1D _k (y));

Among them, y is the feature image before adding attention, C1D represents one-dimensional convolution, σ represents the Sigmoid activation function, and output is the output image after adding attention.

The following table lists the detailed parameters of the convolutional neural network:

Table 1 Detailed parameters of convolutional neural network

The traditional error prediction technology uses several adjacent pixels to predict the target pixel, and cannot well consider the correlation between adjacent pixels. In this embodiment, the pixels in the pixel area are used to predict the target pixel, assuming that the target pixel is ui _,j , and the surrounding pixels vi _,j of the target pixel are used.

All pixels of the original image are divided into two groups: the "cross" set and the "dot" set. In the first stage, the cross set is used for prediction and the point set is used for information embedding; in the second stage, the point set is used for prediction and the cross set is used for information embedding. Therefore, the secret information can be embedded into the difference information including the first interpolation information and the second difference information in the following manner:

The target pixel ui _{,j is predicted by the pixel v i,j in} the area, and the predicted pixel can be expressed as: u' _i,j =CNN(vi _,j );

Obtain d i,j by making the difference between the predicted pixels u' _i,j and u _{i ,j} : d _i,j =u' _i,j -u _i,j ;

Information embedding process: D _i,j =2d _i,j +b;

Among them, D _i,j is the prediction error after expansion, b is the payload to be embedded, and then D _i,j is changed according to the histogram translation method; after the information is embedded, u _i,j is calculated as U _i,j : U _i,j =D _i,j +u'_i,j;

Using the predicted value u' _i,j and the modified pixel value U _i,j , the modified prediction error value is calculated by the following formula: D _i,j =U _i,j -u'_i,j;

The value of the secret information is: b=D _i,j mod2;

The original pixel value is:

As an optional implementation manner, information extraction is performed on the reversible information hiding image in the following manner:

Divide the reversible information hiding image into independent secret-carrying point set images and secret-carrying fork-set images; obtain the predicted secret-carrying fork-set images according to the secret-carrying point set images and fork-set image predictors; determine the secret-carrying fork set images and a second carrier difference image between the predicted fork set images, and secret information and fork set images are extracted from the second carrier difference image;

According to the cross set image and the point set image predictor, obtain the predicted secret point set image; determine the first secret point set image and the predicted secret point set image between the first secret point set image Extract secret information and point set image from secret difference image;

The original image is obtained by fusing the cross set image and the point set image.

In the specific implementation, as shown in Figure 5, the information extraction is the inverse transformation of the embedding process, and the information extraction process is as follows:

(1) At the information extraction end, the secret-carrying image is divided into chessboard-shaped "point" sets and "fork" sets, respectively.

(2) Step 1: Embed the “fork” set in the second stage by using the “point” set and the convolutional neural network model in the training process to predict and embed the “fork” set of the second stage, and combine the predicted “fork” set with the secret “fork” set The difference image M2 is obtained by making a difference, and the secret information S2 and the original difference image are extracted from the difference image M2 at the same time; the extraction process in the second step is the same as the operation in the first step.

(3) The secret-carrying image Im obtains the secret information and the original image I after extracting the secret information in step (2).

Compared with traditional error prediction predictors such as MED, the convolutional deep network error prediction under the same image not only expands the scope of the receptive field in space, but also utilizes the correlation between pixels more efficiently. The attention mechanism is added to the channels of the neural network to enhance the correlation between pixels between channels, improve the prediction performance of the convolutional neural network, and increase the embedding capacity of information.

Selecting MSE, mean, and variance as the evaluation criteria for the prediction performance of the predictor, it can be concluded that the performance based on convolutional neural network is much higher than the traditional prediction performance.

The convolutional neural network error prediction and the traditional MED predictor are compared in the prediction performance of the Lena image, and the comparison results are shown in the following table:

Table 2 Prediction performance comparison

Convolutional Neural Network Predictor Traditional MED Predictor MSE 7.8721 50.7138 mean -0.016 -0.0384 variance 7.8718 50.7125

Moreover, as shown in Figure 6-8, the two-time error prediction peak point of the convolutional neural network is much higher than the MED predictor's error prediction image peak point on the Lena image and the difference image peak point of the target pixel.

Embodiment 2

Please refer to FIG. 9. FIG. 9 is a crowd evacuation congestion propagation prediction system provided by an embodiment of the present application. As shown in FIG. 9, an embodiment of the present application also provides an image reversible information hiding system 900 based on a convolutional neural network, include:

an image division module 910, configured to divide the original image into mutually independent cross-set images and point-set images;

The first secret image acquisition module 920 is used to extract the image features of the receptive fields of different sizes of the fork set image by using a point set image predictor, and perform channel stacking to obtain a predicted point set image; determine the point set image and the predicted point set image. The first difference information between the point set images, the secret information is embedded in the first difference information, and added with the predicted point set image to obtain the first secret image;

The second secret image acquisition module 930 uses a fork set image predictor to extract the image features of the receptive fields of different sizes of the first secret image, and performs channel stacking to obtain a predicted fork set image; determine the fork set image and the prediction the second difference information between the fork set images, the secret information is embedded in the second difference information, and added with the predicted fork set image to obtain a second secret-carrying image;

The fusion module 940 is configured to fuse the first secret-carrying image and the second secret-carrying image to obtain a reversible information hiding image.

Embodiment 3

Please refer to FIG. 10 , which is a schematic diagram of a computer device according to an embodiment of the present application. As shown in FIG. 10 , the computer device 1000 includes a processor 1010 , a memory 1020 and a bus 1030 .

The memory 1020 stores machine-readable instructions executable by the processor 1010. When the computer device 1000 is running, the processor 1010 communicates with the memory 1020 through the bus 1030, and the machine-readable instructions are executed by the bus 1030. When the processor 1010 executes, it can execute the steps of the image reversible information hiding method based on the convolutional neural network in the method embodiments shown in FIG. 1 to FIG. Repeat.

Embodiment 4

Based on the same inventive concept, an embodiment of the present application further provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and the computer program is executed by a processor to execute the methods described in the above method embodiments. Steps of an image reversible information hiding method based on convolutional neural networks.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. During execution, the processes of the embodiments of the above-mentioned methods may be included. The storage medium may be a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM) or the like.

The above descriptions are only preferred embodiments of the present application, and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this application shall be included within the protection scope of this application.

Claims (10)

1. An image reversible information hiding method based on a convolutional neural network is characterized by comprising the following steps:

dividing an original image into a cross set image and a point set image which are independent of each other;

extracting image characteristics of different size receptive fields of the cross set image by using a point set image predictor, and performing channel superposition to obtain a predicted point set image; determining first difference information between the point set image and the predicted point set image, embedding secret information into the first difference information, and adding the first difference information and the predicted point set image to obtain a first secret carrying image;

extracting image characteristics of different sizes of receptive fields of the first secret-carrying image by using a cross set image predictor, and performing channel superposition to obtain a predicted cross set image; determining second difference information between the cross set image and the predicted cross set image, embedding the secret information into the second difference information, and adding the second difference information and the predicted cross set image to obtain a second secret carrying image;

and fusing the first secret-carrying image and the second secret-carrying image to obtain a reversible information hidden image.

2. The reversible information hiding method for images based on the convolutional neural network as claimed in claim 1, wherein the point set image predictor and the cross set image predictor have the same structure and both adopt the convolutional neural network; respectively extracting image features of a target image by using convolution kernels of different sizes of a convolution neural network, and enlarging a receptive field according to superposition combination of the image features, wherein the target image comprises a cross set image and a point set image.

3. The image reversible information hiding method based on the convolutional neural network as claimed in claim 2, wherein the convolutional neural network comprises a trunk prediction network and an auxiliary prediction network, in the trunk prediction network, a target image passes through the convolutional network with an Incept i on structure to output image features of multiple receptive fields with different sizes, channel dimension superposition is performed on the image features, and target pixel prediction is performed by using the image features on different channels to obtain a trunk prediction image; in an auxiliary prediction network, a target image is subjected to prediction of target pixels through a convolution kernel with a specific size, and an auxiliary prediction image is obtained through dimensionality reduction; and determining the predicted image according to the main prediction image and the auxiliary prediction image.

4. The image reversible information hiding method based on the convolutional neural network as claimed in claim 3, wherein a channel attention mechanism is added after the trunk prediction network performs channel dimension superposition on image features to perform weight adjustment on the image features on different channels.

5. The convolutional neural network-based image invertible information hiding method of claim 4, wherein the channel attention mechanism comprises ECA-Net; and after the ECA-Net uses GAP aggregated convolution characteristics without dimension reduction, the size of a one-dimensional convolution kernel is selected in a self-adaptive mode so as to learn channel attention.

6. The convolutional neural network-based image invertible information hiding method according to claim 1, wherein the secret information is embedded into the difference information by:

by pixels v in the area _i,j To predict the target pixel u _i,j The predicted pixel can be expressed as: u. u _i ' _,j ＝CNN(v _i,j )；

By predicting the pixel u _i ' _,j And u _i,j The difference between the two is obtained as d _i,j ：d _i,j ＝u _i ' _,j -u _i,j ；

And (3) information embedding process: d _i,j ＝2d _i,j +b；

Wherein D is _i,j For extended prediction error, b denotes the payload to be embedded, and then D _i,j Changing according to a histogram translation method; after the information is embedded, u _i,j Is calculated as U _i,j :U _i,j ＝D _i,j +u _i ' _,j ；

Using the predicted value u _i ' _,j And modified pixel value U _i,j The modified prediction error value is calculated by: d _i,j ＝U _i,j -u _i ' _,j ；

The value of the secret information is: b is ═ D _i,j mod2；

The values of the original pixels are:

7. the reversible information hiding method for images based on convolutional neural network as claimed in claim 1, characterized in that the reversible information hiding image is subjected to information extraction by:

dividing the reversible information hidden image into a secret-carrying point set image and a secret-carrying cross set image which are independent of each other; obtaining a predicted secret-carrying cross set image according to the secret-carrying point set image and the cross set image predictor; determining a second carrier secret difference image between the carrier secret cross set image and the predicted carrier secret cross set image, and extracting secret information and the cross set image from the second carrier secret difference image;

acquiring a predicted secret point set carrying image according to the fork set image and the point set image predictor; determining a first secret carrying difference image between a secret carrying point set image and a predicted secret carrying point set image, and extracting secret information and a point set image from the first secret carrying difference image;

and fusing the cross set image and the point set image to obtain an original image.

8. An image reversible information hiding system based on a convolutional neural network, comprising:

the image dividing module is used for dividing the original image into a cross set image and a point set image which are independent of each other;

the first secret-carrying image acquisition module is used for extracting image characteristics of different size receptive fields of the cross set image by using the point set image predictor and carrying out channel superposition to obtain a predicted point set image; determining first difference information between the point set image and the predicted point set image, embedding secret information into the first difference information, and adding the first difference information and the predicted point set image to obtain a first secret carrying image;

the second secret-carrying image acquisition module is used for extracting image characteristics of different-size reception fields of the first secret-carrying image by using the cross set image predictor and carrying out channel superposition to obtain a predicted cross set image; determining second difference information between the cross set image and the predicted cross set image, embedding the secret information into the second difference information, and adding the second difference information and the predicted cross set image to obtain a second secret carrying image;

and the fusion module is used for fusing the first secret-carrying image and the second secret-carrying image to obtain a reversible information hidden image.

9. A computer device, comprising: a processor, a memory and a bus, the memory storing machine readable instructions executable by the processor, the processor and the memory communicating via the bus when a computer device is running, the machine readable instructions when executed by the processor performing the steps of the convolutional neural network-based image invertible information hiding method as claimed in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored thereon a computer program which, when being executed by a processor, performs the steps of the convolutional neural network-based image invertible information hiding method as recited in any one of claims 1 to 7.

Images