CN110473135B - Image processing method, system, readable storage medium and intelligent device - Google Patents

Abstract

The invention discloses an image processing method, system, readable storage medium and intelligent equipment. The method includes: acquiring a target mark image, and interfering with the target mark image through an adversarial generation network model to obtain at least one disturbance image; the interference image is converged through the convergence model, and the interference image that is qualified for convergence is used as an adversarial sample, and the adversarial sample is covered on the target mark image; image recognition is performed on the target mark image covered by the adversarial sample, to The wrongly marked image is obtained, and the wrongly marked picture obtained by each user is different; the target mark image of the wrongly marked image is determined according to the matching degree between the wrongly marked image and the target area in the current target marked image. The present invention can solve the problem that the existing watermark is easy to be identified and easily interfered, improves the user's image privacy experience satisfaction, and meets the practical application requirements.

Description

Image processing method, system, readable storage medium and intelligent device

technical field

The present invention relates to the technical field of digital information security, in particular to an image processing method, system, readable storage medium and intelligent equipment.

Background technique

With the continuous development of photography and image processing technology, augmented reality technology has gradually matured. Augmented reality technology is a technology that calculates the position and angle of camera images in real time and adds corresponding images to realize virtual reality on the captured pictures. A high-tech technology in which objects are rendered and interacted with in the real world is widely used in daily life, and image assets are also applied.

Different from content-insensitive image assets (such as news photos, etc.), for content-sensitive image assets, such as private photos, comic works, secret documents, etc., because copyright owners are usually unwilling or need to charge for private disclosure to specified Licensees, whose copyright infringers usually do not reveal their identities when publishing images, therefore detecting and identifying such infringers is an urgent problem. In order to detect and identify such infringers, the existing technology usually adds diverse watermarks for different licensees on the pictures, that is, for different licensees, the watermarks attached to the pictures they receive are all different. same. When an unauthorized infringing picture is found, the identity of the corresponding infringer can be detected by extracting the watermark on the picture, and a lawsuit or compensation can be filed.

However, the existing technology usually faces the two problems of easy identification and interference of the watermark. If the watermark can be recognized by the human eye, it will affect the user's perception of the picture to a certain extent. For example, if a QR code is added to a key position of a picture, the user experience will be severely degraded because the user will see the QR code covering the key position of the picture when viewing the picture. Furthermore, since the failure to extract the watermark correctly would invalidate the process of detecting and identifying the infringer, no action or compensation can be brought.

Contents of the invention

In order to solve the above problems, the purpose of the present invention is to provide an image processing method, system, and readable storage medium that can improve the security of image watermarks and have anti-tampering functions.

The image processing method provided according to the present invention comprises:

Acquiring a target marked image, and interfering with the target marked image through an adversarial generative network model to obtain at least one disturbed image;

Converging the interference image through a convergence model, using the interference image that converges qualified as an adversarial example, and covering the adversarial example on the target mark image;

Perform image recognition on the target labeled image covered by the adversarial example to obtain the wrongly labeled image, and the wrongly labeled images obtained by each user are different;

The target marker image of the wrong marker image is determined according to the matching degree between the wrong marker image and the target area in the current target marker image.

According to the image processing method provided by the present invention, firstly, a target mark image is acquired, and the target mark image is interfered by an adversarial generation network model to obtain at least one interference image; the interference image is converged by a convergence model, and the convergent The qualified interference image is used as an adversarial sample, and the adversarial sample is covered on the target marked image; image recognition is performed on the target marked image covered by the adversarial sample to obtain a wrongly marked image, and the wrongly marked pictures obtained by each user are Not the same; determine the target marker image of the wrong marker image according to the matching degree between the wrong marker image and the target area in the current target marker image. The image processing method provided by the present invention creates different watermarks for each user. Since the generated watermarks cannot be recognized by the human eye, it will not affect the user's perception of the picture, and the algorithm will not be affected when the type of the image recognition model used is uncertain. Unrecognized, so the recognizable area cannot be located. Even if the identifiable area is located, since the added watermark is the result of targeted adversarial training for a specific area, it is difficult for the algorithm to identify the additional information in it. In addition, due to the characteristics of the adversarial samples and the design of multiple identifiable areas, even if the image is defaced, its Watermarks can still be identified and matched to accurately match infringing users.

In addition, according to the above-mentioned image processing method of the present invention, it can also have the following additional technical features:

Further, the step of obtaining a target marked image, and interfering with the target marked image through an adversarial generation network model to obtain at least one disturbed image includes:

acquiring image information of the original image, and discretizing the original image according to the image information, thereby dividing the original image into several identifiable regions;

Marking several identifiable areas of the original image by an image recognition model to obtain the target marked image, the target marked image carries the correct label of the identifiable area;

Interference is performed on the target label image by using an adversarial generative network model to obtain a corresponding interference image, and the interference image carries an erroneous label of an identifiable region.

Further, the step of marking several identifiable regions of the original image through an image recognition model to obtain the target marked image includes:

Extracting and decomposing several identifiable regions of the original image to form marking factors;

The correct label of the identifiable region is obtained after machine algorithm learning is performed according to the labeling factor.

Further, the formula for calculating the correct label of the identifiable region after machine algorithm learning according to the labeling factor is:

Among them, φ() is the activation function, V _k is the adjustment coefficient, W _i is the initial weight, Xi is the marking factor, and _{B k is} the offset accumulation.

Further, the wrongly marked image corresponds to an initial weight matrix, and the step of determining the target mark image of the wrongly marked image according to the matching degree between the wrongly marked image and the target area in the current target mark image includes:

calculating a correlation matrix between the wrongly marked image and the current target marked image;

Modify the initial weight matrix according to the correlation matrix to obtain a target weight matrix, and the target weight matrix is used to adjust the initial weight when the wrongly marked image matches the current target marked image Matrix to replace;

judging the degree of matching between the error-marked image corresponding to the replacement of the initial weight matrix and the current target-marked image;

When the matching degree between the wrongly marked image and the current target marked image is greater than a preset matching value, the current target marked image is used as the target marked image of the wrongly marked image.

Further, the preset matching value ranges from 93% to 98%.

Further, after the step of judging the matching degree between the corresponding wrongly marked image after replacing the initial weight matrix and the current target marked image, the method further includes:

When the matching degree between the wrongly marked image and the current target marked image is less than a preset matching value, it is judged whether the wrongly marked image carries the associated identification of the current target marked image;

If so, querying the associated matching library according to the associated identification, and matching the associated matching library with the wrongly marked image to obtain associated data, and determining the target tagged image of the incorrectly marked image according to the associated data;

If not, an alarm prompt is issued.

Another embodiment of the present invention proposes an image processing system, which solves the problems of easy recognition and interference of existing watermarks, and improves user satisfaction with image privacy experience.

An image processing system according to an embodiment of the present invention includes:

A determining module, configured to determine whether the current source address is a cold source address according to the space access amount of the current source address within the access period;

A judging module, configured to judge whether the reference count of the data block at the first access location of the current source address is less than a preset value when the current source address is a cold source address;

A deletion module, configured to delete the current source address;

A migration module, configured to move and store the current source address to a second access location, where the reference count of the data block at the second access location is greater than that at the first access location.

Another embodiment of the present invention also proposes a storage medium on which a computer program is stored, and when the program is executed by a processor, the steps of the above method are implemented.

Another embodiment of the present invention also proposes an intelligent device, including a memory, a processor, and a computer program stored in the memory and operable on the processor, and the processor implements the steps of the above method when executing the program.

Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be apparent from the description which follows, or may be learned by practice of the invention.

Description of drawings

Fig. 1 is the flowchart of the image processing method that the first embodiment of the present invention proposes;

Fig. 2 is the specific flowchart of step S101 in Fig. 1;

Fig. 3 is the specific flowchart of step S104 in Fig. 1;

FIG. 4 is a structural block diagram of an image processing system proposed by a second embodiment of the present invention.

detailed description

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments It is a part of embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

Please refer to FIG. 1, an image processing method proposed in the first embodiment of the present invention, which includes steps S101 to S104:

Step S101 , acquiring a target marked image, and interfering with the target marked image through an adversarial generative network model to obtain at least one disturbed image.

In this embodiment, an image processing device is taken as an example for illustration, but it should be understood that the embodiment of the present invention is not limited thereto, and the method of the embodiment of the present invention can be applied to any smart device, that is, any device capable of image processing in electronic equipment. Specifically, in the prior art, multiple watermarks for different authorized persons are usually added to the pictures, that is, for different authorized persons, the watermarks attached to the pictures they receive are all different. When an unauthorized infringing picture is found, the identity of the corresponding infringer can be detected by extracting the watermark on the picture, and a lawsuit or compensation can be filed. However, the watermark is easily recognized and easily disturbed. Therefore, if the watermark can be recognized by the human eye, it will affect the user's perception of the picture to a certain extent. If the watermark is easy to remove or interfere, then when the infringing picture is made public, the watermark cannot be extracted correctly, which will lead to the detection and identification of infringement. The process of lapse of the identity of the owner makes it impossible to sue or indemnify for it.

In this embodiment, when a target image acquisition instruction is received; the target marked image is acquired, and the target marked image is interfered with by an adversarial generation network model to obtain at least one disturbed image, so that unauthorized users can When an image is made public, its identity can be traced to bring a lawsuit or compensation. In addition, it can prevent unauthorized users from correctly identifying the image information of the image, which improves the security of the image.

Please refer to FIG. 2 , to obtain a target mark image, and to interfere with the target mark image through the confrontation generation network model, so as to obtain at least one interference image. The method includes the following steps:

Step S1011, acquiring image information of the original image, and performing discretization processing on the original image according to the image information, so as to divide the original image into several identifiable regions.

Step S1012, using an image recognition model to mark several identifiable regions of the original image to obtain the target marked image, and the target marked image carries the correct label of the identifiable region.

Specifically, several identifiable regions of the original image are extracted and decomposed to form marking factors; machine algorithm learning is performed according to the marking factors to obtain correct labels of identifiable regions.

The formula for calculating the correct label of the identifiable region after machine algorithm learning according to the labeling factor is:

Among them, φ() is the activation function, V _k is the adjustment coefficient, W _i is the initial weight, Xi is the marking factor, and _{B k is} the offset accumulation.

As mentioned above, by inputting the image information of the original image, after data standardization and extraction decomposition, the marking factor X _i of the original image is obtained, and the corresponding initial weight W _i is configured, and the summation and B _k offset accumulation are carried out, and the V _k coefficient Conditioning, the activation function φ(), outputs the correct labels for recognizable regions.

In step S1013, the object label image is interfered with by using the confrontation generative network model to obtain a corresponding interference image, and the interference image carries the wrong label of the identifiable region.

As mentioned above, by discretizing the original image, the original image is divided into several identifiable areas, rather than random selection, fixed positions, the entire picture, or electronic information, so that users can identify the identifiable areas based on the As a result, the corresponding target mark image is determined. Each picture can support m ⁿ users, where m is the number of identifiable areas in the picture, and n is the number of marks identifiable by the image recognition model, and the size of each identifiable area can be consistent. Can be inconsistent; several identifiable regions of the original image are marked by an image recognition model (such as YOLOv3) to obtain the target marker image, wherein the target marker image carries the correct label of the identifiable region, and the correct label covers For the entire original image, for example, identifiable area 1 is identified as label a, identifiable area 2 is identified as label b, and identifiable area 3 is identified as label c.

Further, the target marker image is interfered with by an adversarial generative network model to obtain a corresponding interference image, which carries an erroneous label of an identifiable region. It can be understood that the erroneous label is generated by a recognition error, This recognition error is called the difference value (loss). By setting different training objectives, the confrontation generative network model can generate different interference images and produce different difference values until it converges and obtains a stable recognition error. For example, when the confrontation generation network converges through the specified convergence model (such as BP-Gradient), the interference image generated by it can make the image recognition model accurately and stably identify the identifiable area 1 as the label x, and the identifiable area 2 is wrong is identified as label y, and the identifiable area 3 is incorrectly identified as label z.

In step S102, the interference image is converged through a convergence model, and the interference image that is qualified for convergence is used as an adversarial example, and the adversarial example is overlaid on the target marked image.

As mentioned above, when the adversarial generation network converges through the specified convergence model (such as BP-Gradient), the generated interference image is called an adversarial sample, and the adversarial sample is overlaid on the current target image to facilitate the current Even if the image is defaced, its watermark can still be identified and matched, and the infringing user can be accurately matched. Due to the generation characteristics of adversarial samples, they are usually pixel perturbations, which are difficult for human eyes to recognize, and through different recognition errors, their sequences and positions can be embedded with different user identification information. For example: the image obtained by user A is the image covered by the adversarial samples that mistakenly identify the identifiable areas 1, 2, and 3 as x, y, and z respectively; the image obtained by user B is the image obtained by identifiable areas 1, 2 and 3 are respectively wrongly identified as the images covered by the adversarial examples of i, j, and k. The pictures obtained by user A and user B respectively carry different information that can indicate their identities, so that the owner of the picture can be found.

Step S103 , image recognition is performed on the target labeled image covered by the adversarial example to obtain an incorrectly labeled image, wherein the incorrectly labeled images obtained by each user are different.

Specifically, the image recognition model performs image recognition on the original image covered by the adversarial example, and can obtain the target error label corresponding to the identifiable area of the image. The identifiable area 1 is identified as label x, and the identifiable area 2 is identified as Label y and recognizable area 3 are recognized as label z, and its "wrong label" specifically refers to the label marked by the specified image recognition model that does not match the label marked in the original image. Since each user gets a different mislabeled picture, the owner of the picture can be traced.

Step S104, determining the target marker image of the wrong marker image according to the matching degree between the wrong marker image and the target area in the current target marker image.

Please refer to FIG. 3 , the wrongly marked image corresponds to an initial weight matrix, and the step of determining the target mark image of the wrongly marked image according to the degree of matching between the wrongly marked image and the target region in the current target mark image includes:

Step S1041, calculating a correlation matrix between the wrongly marked image and the current target marked image.

Step S1042, correcting the initial weight matrix according to the correlation matrix to obtain a target weight matrix, and the target weight matrix is used for matching the wrongly marked image with the current target marked image. The above initial weight matrix is replaced.

Step S1043 , judging the matching degree between the error mark image corresponding to the replacement of the initial weight matrix and the current target mark image.

Step S1044, when the matching degree between the wrongly marked image and the current target marked image is greater than a preset matching value, use the current target marked image as the target marked image of the wrongly marked image. The value range of the preset matching value is 93%-98%.

As mentioned above, when matching and comparing the wrongly marked image with the current target marked image, first calculate the correlation matrix M(i, j) between the wrongly marked image and the current target marked image, and according to the obtained correlation matrix M(i,j) modifies the initial weight matrix to obtain a target weight matrix K'(i,j), after determining the target weight matrix K'(i,j), judge the initial weight matrix K(i, j) is the matching degree between the corresponding wrongly labeled image and the current target labeled image after replacement. Since there is a difference in the correlation value corresponding to the correlation matrix, which directly reflects the matching degree between the wrongly marked image and the current target marked image, the initial weight The matrix K(i,j) is corrected to obtain the target weight matrix K'(i,j). When performing actual matching, replace the original initial weight matrix with the target weight matrix K'(i,j), so that the value of the replaced target weight matrix K'(i,j) is the same as the value of the above-mentioned correlation matrix It is more consistent, thereby improving the corresponding matching accuracy when matching each other. When the matching degree between the wrongly marked image and the current target marked image is greater than a preset matching value of 93%-98%, the current target marked image is used as the target marked image of the wrongly marked image.

It should also be noted here that after the step of judging the matching degree between the wrongly marked image corresponding to the replacement of the initial weight matrix and the current target marked image, the method further includes: when the wrongly marked image and When the matching degree between the current target mark images is less than the preset matching value, it is judged whether the wrong mark image carries the association mark of the current target mark image; The matching library is matched with the wrongly marked image to obtain associated data, and the target marked image of the wrongly marked image is determined according to the associated data; if not, an alarm prompt is issued.

As mentioned above, by judging whether the erroneous tagged image carries the associated identifier of the current target tagged image, when the erroneously tagged image carries the associated identifier of the current target tagged image, query the associated matching library according to the associated identifier, and The association matching library is matched with the wrongly marked image to obtain associated data, so as to determine the target marked image of the wrongly marked image according to the associated data, which is convenient for users to verify image information through a multi-party platform. In addition, through The reminder of the alarm information is convenient for users to protect their own rights and interests in a timely manner, and has a certain deterrent effect.

As a specific example, since different error labeling information corresponds to different users, for example, the labels of image A obtained by image recognition of the original image covered by the adversarial example through the image recognition model are x, y, z; and the image B uses the image recognition model to perform image recognition on the original image covered by the adversarial example, and the labels obtained are i, j, and k. Then it can be known that image A belongs to user A and image B belongs to user B through database matching. In addition, since the information sequence can be embedded with certain anti-jamming properties, even if the image recognition model cannot correctly identify the labels of some recognizable areas due to image defacement, the result can still match the corresponding user. For example, even if image A is labeled as x, y by image recognition of the original image covered by the adversarial example through the image recognition model, but the recognizable area 3 is not correctly identified, because x, y is related to i, j, k The former is more similar to user A's labels x, y, and z, and it can still be concluded that picture A belongs to user A.

According to the image processing method provided by the present invention, firstly, a target mark image is acquired, and the target mark image is interfered by an adversarial generation network model to obtain at least one interference image; the interference image is converged by a convergence model, and the convergent The qualified interference image is used as an adversarial sample, and the adversarial sample is covered on the target marked image; image recognition is performed on the target marked image covered by the adversarial sample to obtain a wrongly marked image, and the wrongly marked pictures obtained by each user are Not the same; determine the target marker image of the wrong marker image according to the matching degree between the wrong marker image and the target area in the current target marker image. The image processing method provided by the present invention creates different watermarks for each user. Since the generated watermarks cannot be recognized by the human eye, it will not affect the user's perception of the picture, and the algorithm will not be affected when the type of the image recognition model used is uncertain. Unrecognized, so the recognizable area cannot be located. Even if the identifiable area is located, since the added watermark is the result of targeted adversarial training for a specific area, it is difficult for the algorithm to identify the additional information in it. In addition, due to the characteristics of the adversarial samples and the design of multiple identifiable areas, even if the image is defaced, its Watermarks can still be identified and matched to accurately match infringing users.

Please refer to Fig. 4, based on the same inventive concept, the image processing system provided by the second embodiment of the present invention includes:

The acquiring module 10 is configured to acquire an image of a target mark, and interfere with the image of the target mark by using an adversarial generative network model to obtain at least one interference image.

In this embodiment, the acquisition module 10 includes:

The acquiring unit 11 is configured to acquire image information of the original image, and perform discretization processing on the original image according to the image information, so as to divide the original image into several identifiable regions.

The marking unit 12 is configured to mark several identifiable areas of the original image through an image recognition model to obtain the target marked image, and the target marked image carries the correct label of the identifiable area.

Specifically, several identifiable regions of the original image are extracted and decomposed to form marking factors; machine algorithm learning is performed according to the marking factors to obtain correct labels of identifiable regions.

The formula for calculating the correct label of the identifiable region after machine algorithm learning according to the labeling factor is:

Among them, φ() is the activation function, V _k is the adjustment coefficient, W _i is the initial weight, Xi is the marking factor, and _{B k is} the offset accumulation.

The interference unit 13 is configured to interfere with the target marker image by using an adversarial generative network model to obtain a corresponding interference image, and the interference image carries an erroneous label of an identifiable region.

The convergence module 20 is configured to converge the interference image through a convergence model, use the interference image that has converged qualified as an adversarial example, and cover the adversarial example on the target marked image.

The recognition module 30 is configured to perform image recognition on the target marked image covered by the adversarial example to obtain the wrongly marked image, and the wrongly marked pictures obtained by each user are different.

The determination module 40 is configured to determine the target mark image of the wrong mark image according to the matching degree between the wrong mark image and the target area in the current target mark image.

In this embodiment, the determination module 40 includes:

The calculation unit 41 is configured to calculate a correlation matrix between the wrongly marked image and the current target marked image.

A correction unit 42, configured to correct the initial weight matrix according to the correlation matrix to obtain a target weight matrix, and the target weight matrix is used when the wrongly marked image is matched with the current target marked image , to replace the initial weight matrix.

The judging unit 43 is configured to judge the matching degree between the error mark image corresponding to the replacement of the initial weight matrix and the current target mark image.

The determining unit 44 is configured to use the current target marker image as the target marker image of the wrong marker image when the matching degree between the wrong marker image and the current target marker image is greater than a preset matching value. The value range of the preset matching value is 93%-98%.

The judging unit 43 is further configured to determine whether the wrongly marked image carries an association identifier of the current target marked image when the matching degree between the wrongly marked image and the current target marked image is less than a preset matching value; if so, Then query the associated matching library according to the associated identification, and match the associated matching library with the wrongly marked image to obtain associated data, and determine the target tagged image of the incorrectly marked image according to the associated data; if not , an alarm will be issued.

According to the image processing system provided by the present invention, firstly, a target mark image is obtained, and the target mark image is interfered with by an adversarial generation network model to obtain at least one interference image; the convergence model is used to converge the interference image, and the converged The qualified interference image is used as an adversarial sample, and the adversarial sample is covered on the target marked image; image recognition is performed on the target marked image covered by the adversarial sample to obtain a wrongly marked image, and the wrongly marked pictures obtained by each user are Not the same; determine the target marker image of the wrong marker image according to the matching degree between the wrong marker image and the target area in the current target marker image. The image processing method provided by the present invention creates different watermarks for each user. Since the generated watermarks cannot be recognized by the human eye, it will not affect the user's perception of the picture, and the algorithm will not be affected when the type of the image recognition model used is uncertain. Unrecognized, so the recognizable area cannot be located. Even if the identifiable area is located, since the added watermark is the result of targeted adversarial training for a specific area, it is difficult for the algorithm to identify the additional information in it. In addition, due to the characteristics of the adversarial samples and the design of multiple identifiable areas, even if the image is defaced, its Watermarks can still be identified and matched to accurately match infringing users.

The technical features and technical effects of the image processing system proposed in the embodiment of the present invention are the same as those of the method proposed in the embodiment of the present invention, and will not be repeated here.

In addition, an embodiment of the present invention also proposes a storage medium on which a computer program is stored, and when the program is executed by a processor, the steps of the above method are implemented.

In addition, an embodiment of the present invention also proposes an intelligent device, including a memory, a processor, and a computer program stored in the memory and operable on the processor, wherein the processor implements the above method when executing the program step.

The logic and/or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced listing of executable instructions for implementing logical functions, can be embodied in any computer-readable medium, For use with instruction execution systems, devices, or devices (such as computer-based systems, systems including processors, or other systems that can fetch instructions from instruction execution systems, devices, or devices and execute instructions), or in conjunction with these instruction execution systems, devices or equipment for use. For the purposes of this specification, a "computer-readable medium" may be any device that can contain, store, communicate, propagate or transmit a program for use in or in conjunction with an instruction execution system, device or device.

More specific examples (non-exhaustive list) of computer-readable media include the following: electrical connection with one or more wires (electronic device), portable computer disk case (magnetic device), random access memory (RAM), Read Only Memory (ROM), Erasable and Editable Read Only Memory (EPROM or Flash Memory), Fiber Optic Devices, and Portable Compact Disc Read Only Memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program can be printed, since the program can be read, for example, by optically scanning the paper or other medium, followed by editing, interpretation or other suitable processing if necessary. processing to obtain the program electronically and store it in computer memory.

It should be understood that various parts of the present invention can be realized by hardware, software, firmware or their combination. In the above described embodiments, various steps or methods may be implemented by software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented by any one or combination of the following techniques known in the art: Discrete logic circuits, ASICs with suitable combinational logic gates, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

In the description of this specification, descriptions referring to the terms "one embodiment", "some embodiments", "example", "specific examples", or "some examples" mean that specific features described in connection with the embodiment or example , structure, material or characteristic is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

Although the embodiments of the present invention have been shown and described, those skilled in the art can understand that various changes, modifications, substitutions and modifications can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the invention is defined by the claims and their equivalents.

Finally, it should be noted that: the above-described embodiments are only specific implementations of the present invention, to illustrate the technical solutions of the present invention, rather than to limit it, and the scope of protection of the present invention is not limited thereto, although referring to the foregoing The embodiment has described the present invention in detail, and those of ordinary skill in the art should understand that any person familiar with the technical field can still modify the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention Changes can be easily thought of, or equivalent replacements are made to some of the technical features; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be covered by the scope of the present invention. within the scope of protection. Therefore, the protection scope of the present invention should be determined by the protection scope of the claims.

Claims (10)

1. An image processing method, characterized in that it comprises the steps of:

obtaining a target mark image, and interfering the target mark image through a countermeasure generation network model to obtain at least one interference image;

converging the interference image through a convergence model, taking the interference image qualified in convergence as a countermeasure sample, and covering the countermeasure sample on a target mark image;

carrying out image recognition on the target marked image covered by the countermeasure sample to obtain error marked images, wherein the error marked images obtained by each user are different;

and determining the target marking image of the error marking image according to the matching degree of the error marking image and the target area in the current target marking image.

2. The image processing method of claim 1, wherein the step of obtaining a target marker image, and disturbing the target marker image by countering the generative network model to obtain at least one disturbed image comprises:

acquiring image information of an original image, and performing discretization processing on the original image according to the image information so as to divide the original image into a plurality of recognizable areas;

marking a plurality of identifiable regions of the original image through an image identification model to obtain a target marking image, wherein the target marking image carries correct labels of the identifiable regions;

and interfering the target mark image by a counter-generation network model to obtain a corresponding interference image, wherein the interference image carries an error label of an identifiable region.

3. The image processing method according to claim 2, wherein the step of marking a plurality of recognizable areas of the original image by an image recognition model to obtain the target mark image comprises:

extracting and decomposing a plurality of identifiable regions of the original image to form marking factors;

and obtaining the correct label of the identifiable region after machine algorithm learning according to the marking factor.

4. The image processing method according to claim 3, wherein the calculation formula of the correct label of the recognizable area obtained by machine algorithm learning according to the labeling factor is as follows:

where φ () is an activation function, V _k To adjust the coefficient, W _i As an initial weight, X _i As a marker factor, B _k Is the offset accumulation.

5. The image processing method according to claim 1, wherein the error marked image corresponds to an initial weight matrix, and the step of determining the target marked image of the error marked image according to the matching degree of the error marked image and the target area in the current target marked image comprises:

calculating a correlation matrix between the error marked image and the current target marked image;

correcting the initial weight matrix according to the correlation matrix to obtain a target weight matrix, wherein the target weight matrix is used for replacing the initial weight matrix when the error marked image is matched with the current target marked image;

judging the matching degree between the corresponding error marked image and the current target marked image after the initial weight matrix is replaced;

and when the matching degree between the error mark image and the current target mark image is greater than a preset matching value, taking the current target mark image as the target mark image of the error mark image.

6. The image processing method according to claim 5, wherein the preset matching value ranges from 93% to 98%.

7. The image processing method according to claim 5, wherein after the step of determining the matching degree between the error marked image corresponding to the replaced initial weight matrix and the current target marked image, the method further comprises:

when the matching degree between the error marked image and the current target marked image is smaller than a preset matching value, judging whether the error marked image carries an associated identifier of the current target marked image;

if yes, inquiring an association matching library according to the association identifier, matching the association matching library with the error marked image to obtain association data, and determining a target marked image of the error marked image according to the association data;

if not, an alarm prompt is sent out.

8. An image processing system, characterized in that the system comprises:

the acquisition module is used for acquiring a target mark image and interfering the target mark image through a countermeasure generation network model to obtain at least one interference image;

the convergence module is used for converging the interference image through a convergence model, taking the interference image qualified in convergence as a countermeasure sample, and covering the countermeasure sample on a target mark image;

the identification module is used for carrying out image identification on the target mark image covered by the countermeasure sample to obtain error mark images, and the error mark images obtained by each user are different;

and the determining module is used for determining the target mark image of the error mark image according to the matching degree of the error mark image and the target area in the current target mark image.

9. A readable storage medium on which a computer program is stored which, when being executed by a processor, carries out the image processing method of any one of claims 1 to 7.

10. An intelligent device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the image processing method of any of claims 1 to 7 when executing the program.

Images