Image detection method and device based on noise enhancement, equipment and storage medium

Abstract

The application discloses an image detection method and device based on noise enhancement, equipment and a storage medium, and belongs to the technical field of image detection. The application inputs a to-be-detected image into a trained image tampering detection model, and performs global noise enhancement processing on the to-be-detected image to obtain a processed to-be-detected image; high-dimensional feature extraction is performed on the processed to-be-detected image to obtain a to-be-detected feature image; local noise enhancement processing is performed on the to-be-detected feature image to obtain an abnormal noise-enhanced to-be-detected feature image; and detection is performed on the abnormal noise-enhanced to-be-detected feature image to obtain an image detection recognition result. The application first performs global noise enhancement on a to-be-detected image based on a trained image tampering detection model, and then performs feature extraction and local noise enhancement on the to-be-detected image to obtain a to-be-detected feature image, and further performs detection on the to-be-detected feature image to obtain a corresponding detection result, thereby improving the detection accuracy of tampered images.

Description

Technical Field

[0001] This invention relates to the field of image detection technology, and in particular to an image detection method, apparatus, device, and storage medium based on noise enhancement. Background Technology

[0002] As images play an increasingly important role in information dissemination, attacks and tampering with images have become a means for malicious actors to achieve their goals. Simultaneously, the rapid development of various image processing tools has lowered the technical barriers to image manipulation and tampering, making malicious image tampering a significant threat to information security. Compared to the few digital images whose importance is clearly defined and watermarked for copyright protection during image generation, the number of watermark-free digital images in modern information dissemination is far greater. On the other hand, watermarked images are easier to detect and protect in terms of their original data and integrity when tampered with. These watermarks allow for self-recovery after tampering, providing reliability in tamper detection. However, tamper detection technology needs to address the threat of a large number of watermark-free images being tampered with and unable to self-recover during digital information dissemination. This requires accurate identification of tampered images and precise location of the tampered area. This technology can be widely applied in various fields and is suitable for various business scenarios requiring non-closed-loop information dissemination.

[0003] Therefore, there is an urgent need for an image detection method based on noise enhancement for image tampering detection.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this invention is to provide an image detection method, apparatus, device, and storage medium based on noise enhancement, aiming to solve the technical problem of low accuracy in detecting tampered images in the prior art.

[0006] To achieve the above objectives, the present invention provides an image detection method based on noise enhancement, the method comprising the following steps:

[0007] The image to be detected is input into the trained image tampering detection model, and global noise enhancement is performed on the image to be detected to obtain the processed image to be detected.

[0008] High-dimensional feature extraction is performed on the processed image to be detected to obtain the feature image to be detected;

[0009] The image of the feature to be detected is subjected to local noise enhancement processing to obtain the image of the feature to be detected after abnormal noise enhancement;

[0010] The image of the feature to be detected after the abnormal noise enhancement is performed to obtain the image detection and recognition result.

[0011] Optionally, the feature image to be detected is subjected to local noise enhancement processing to obtain a feature image to be detected after abnormal noise enhancement. Specific steps include:

[0012] Obtain noise feature image blocks and perform average pooling on the noise feature image blocks to obtain noise convolution kernels;

[0013] The noise convolution kernel is used to traverse the feature image to be detected layer by layer, and feature activation is performed on the common feature image region to be enhanced to obtain the feature image to be detected after abnormal noise enhancement.

[0014] Optionally, noise feature image blocks are obtained, and average pooling is performed on the noise feature image blocks to obtain noise convolution kernels. Specific steps include:

[0015] The processed image to be detected is randomly segmented to obtain multiple noise feature image blocks;

[0016] The noise feature image block is subjected to average pooling to obtain the noise convolution kernel.

[0017] Optionally, the enhanced image of the features to be detected after the abnormal noise is subjected to detection to obtain the image detection and recognition result. Specific steps include:

[0018] Initial image detection is performed on the enhanced image of the feature to be detected after abnormal noise to obtain the first image detection result;

[0019] Depth image detection is performed on the first image detection result to obtain the image detection and recognition result.

[0020] Optionally, depth image detection is performed on the first image detection result to obtain the image detection and recognition result. Specific steps include:

[0021] Global information focus is applied to the detection result of the first image to obtain the global information recognition result;

[0022] Local information focus is applied to the detection result of the first image to obtain the local information recognition result;

[0023] Cross-information attention is applied to the first image detection result to obtain the cross-information recognition result;

[0024] The global information recognition result, the local information recognition result, and the cross-information recognition result are integrated to obtain the image detection and recognition result.

[0025] Optionally, the training process of the trained image tampering detection model includes:

[0026] Obtain the training image dataset and perform data preprocessing on the training image data in the training image dataset to obtain the processed training image dataset;

[0027] Data feature extraction is performed on the processed training image data in the processed training image dataset to obtain the image feature dataset;

[0028] The training image feature dataset is classified to obtain an image feature training set and an image feature test set;

[0029] The image feature training set is imported into the image tampering detection model to be trained for model training, and the image tampering detection model to be detected is obtained.

[0030] The image feature test set is imported into the image tampering detection model to be detected for model detection, and the trained image tampering detection model is output when the model detection result is qualified.

[0031] Optionally, a training image dataset is obtained, and the training image data in the training image dataset is preprocessed to obtain a processed training image dataset. Specific steps include:

[0032] Obtain the training image dataset, and split each training image data in the training image dataset into tampered images to obtain the corresponding tampered part image and multiple untampered part images;

[0033] For each training image data, each tampered image and its corresponding untampered image are bound together to obtain the positive and negative image instances corresponding to each tampered image.

[0034] The obtained positive and negative image instances are integrated to obtain the processed training image dataset.

[0035] Furthermore, to achieve the above objectives, the present invention also proposes an image detection device based on noise enhancement, the image detection device based on noise enhancement comprising:

[0036] Preprocessing module: Input the image to be detected into the trained image tampering detection model, and perform global noise enhancement processing on the image to be detected to obtain the processed image to be detected;

[0037] Feature extraction module: performs high-dimensional feature extraction on the processed image to be detected to obtain the feature image to be detected;

[0038] Noise enhancement module: Performs local noise enhancement processing on the feature image to be detected to obtain the feature image to be detected after abnormal noise enhancement;

[0039] Image detection module: Detects the image of the feature to be detected after the abnormal noise enhancement, and obtains the image detection and recognition result.

[0040] Furthermore, to achieve the above objectives, the present invention also proposes a noise-enhanced image detection device, which includes: a memory, a processor, and a noise-enhanced image detection program stored in the memory and executable on the processor. The noise-enhanced image detection program is configured to implement the steps of the noise-enhanced image detection method described above.

[0041] Furthermore, to achieve the above objectives, the present invention also proposes a computer-readable storage medium storing a computer program, wherein the storage medium stores a noise-enhanced image detection program, and when the noise-enhanced image detection program is executed by a processor, it implements the steps of the noise-enhanced image detection method described above.

[0042] This invention inputs the image to be detected into a trained image tampering detection model and performs global noise enhancement on the image to obtain a processed image. High-dimensional feature extraction is then performed on the processed image to obtain a feature image to be detected. Local noise enhancement is then performed on the feature image to obtain a feature image with anomalously enhanced noise. Finally, the feature image with anomalously enhanced noise is detected to obtain an image detection and recognition result. This invention, based on a trained image tampering detection model, first performs global noise enhancement on the image to be detected, thereby strengthening the noise features of the tampered portion of the image. Then, feature extraction and local noise enhancement are performed to obtain a feature image to be detected. Finally, detection is performed to obtain the corresponding detection result, thus improving the accuracy of tampering image detection. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of the structure of a noise-enhanced image detection device in the hardware operating environment of the embodiment of the present invention;

[0044] Figure 2 This is a flowchart illustrating the first embodiment of the image detection method based on noise enhancement according to the present invention;

[0045] Figure 3 This is a flowchart illustrating the second embodiment of the image detection method based on noise enhancement according to the present invention;

[0046] Figure 4This is a flowchart illustrating the third embodiment of the image detection method based on noise enhancement of the present invention;

[0047] Figure 5 This is a flowchart illustrating the fourth embodiment of the image detection method based on noise enhancement of the present invention;

[0048] Figure 6 This is a structural block diagram of the first embodiment of the image detection device based on noise enhancement according to the present invention.

[0049] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0050] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0051] Reference Figure 1 , Figure 1 This is a schematic diagram of the structure of a noise-enhanced image detection device in the hardware operating environment of an embodiment of the present invention.

[0052] like Figure 1 As shown, the noise-enhanced image detection device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wireless-Fidelity (Wi-Fi) interface). The memory 1005 may be high-speed random access memory (RAM) or stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0053] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on noise-enhanced image detection equipment and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0054] like Figure 1As shown, the memory 1005, which serves as a storage medium, may include an operating system, a network communication module, a user interface module, and a noise-enhanced image detection program.

[0055] exist Figure 1 In the noise-enhanced image detection device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and memory 1005 in the noise-enhanced image detection device of the present invention can be set in the noise-enhanced image detection device, and the noise-enhanced image detection device calls the noise-enhanced image detection program stored in the memory 1005 through the processor 1001 and executes the noise-enhanced image detection method provided in the embodiment of the present invention.

[0056] This invention provides an image detection method based on noise enhancement, referring to... Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of an image detection method based on noise enhancement according to the present invention.

[0057] In this embodiment, the image detection method based on noise enhancement includes the following steps:

[0058] Step S10: Input the image to be detected into the trained image tampering detection model, and perform global noise enhancement processing on the image to be detected to obtain the processed image to be detected;

[0059] It should be noted that, in the specific implementation, the global noise enhancement processing of the image to be detected is achieved through NoiseConv, which is composed of noise kernels such as the SRM kernel, Bayar kernel, and Command kernel, each targeting different types of noise. Specifically, the SRM (Steganalysis Rich Model) kernel is a noise kernel formed by a rich steganalysis model; the Bayar kernel is a noise kernel formed by a Bayar model; the Command kernel is a general color gamut kernel formed by a deep learning model; and NoiseConv is a noise conversion model that integrates noise.

[0060] It should also be noted that in the specific implementation, the final size and number of channels of the input image to be detected remain unchanged. Furthermore, since the noise signal in the tampered area of ​​the image to be detected is stronger than that in the surrounding normal area, after global noise enhancement, the noise texture of the tampered area of ​​the image to be detected will have a more significant difference from the noise texture of the untampered area, thus helping the subsequent feature extraction layer to more effectively capture the tampered features.

[0061] Step S20: Perform high-dimensional feature extraction on the processed image to be detected to obtain the feature image to be detected;

[0062] It should be noted that the high-dimensional feature extraction of the processed image to be detected is specifically achieved by the convolutional layers with different parameter settings in the trained image tampering detection model, which extract high-dimensional features from the entire image to be detected. The parameter settings of each convolutional layer are determined through deep learning during the training of the image tampering detection model.

[0063] Step S30: Perform local noise enhancement processing on the feature image to be detected to obtain the feature image to be detected after abnormal noise enhancement;

[0064] Understandably, in the specific implementation, local noise enhancement is achieved through the feature enhancement module in the trained image tampering detection model, and its specific function is to activate the tampered region of the feature image to be detected.

[0065] It should also be noted that, in the specific implementation, the local noise enhancement process of the feature image to be detected is as follows: first, the global noise enhancement image, i.e. the processed image to be detected, is randomly cut into noise feature image blocks; then, the noise feature image blocks are processed by average pooling to obtain a 1*1*1024 noise convolution kernel; finally, the noise convolution kernel is used to traverse the feature image to be detected layer by layer to find common feature regions and activate them, thereby obtaining the feature image to be detected after abnormal noise enhancement.

[0066] Step S40: Detect the feature image to be detected after the abnormal noise enhancement to obtain the image detection and recognition result.

[0067] It should be noted that the image detection and recognition result is specifically a judgment result on whether the image to be detected has been tampered with. In the specific implementation, the image detection and recognition result is generally divided into two categories: one is that if the image to be detected has not been tampered with, the image detection and recognition result is that the image detection is normal; the other is that if the image to be detected has been tampered with, the image detection and recognition result is that the image detection is abnormal, and the corresponding tampering location information is output.

[0068] This embodiment inputs the image to be detected into a trained image tampering detection model, and performs global noise enhancement on the image to obtain a processed image. High-dimensional feature extraction is then performed on the processed image to obtain a feature image to be detected. Local noise enhancement is then performed on the feature image to obtain a feature image with enhanced abnormal noise. Finally, the enhanced feature image is detected to obtain the image detection and recognition result. This embodiment, based on the trained image tampering detection model, first performs global noise enhancement on the image to be detected, thereby strengthening the noise features of the tampered portion of the image. Then, feature extraction and local noise enhancement are performed to obtain the feature image to be detected. Finally, detection is performed to obtain the corresponding detection result, improving the accuracy of tampering image detection.

[0069] refer to Figure 3 , Figure 3 This is a flowchart illustrating a second embodiment of an image detection method based on noise enhancement according to the present invention.

[0070] Based on the first embodiment described above, in this embodiment, step S30 specifically includes the following steps:

[0071] Step S31: Obtain the noise feature image block and perform average pooling on the noise feature image block to obtain the noise convolution kernel;

[0072] It should be noted that, in the specific implementation, the noise feature image block is obtained by randomly cutting the global noise enhancement map, i.e., the processed image to be detected.

[0073] It should also be noted that average pooling divides the input image into several rectangular regions and outputs the average value of all elements for each sub-region. In this embodiment, specifically, the noisy feature image block is divided into several rectangular regions, and then the average value of all elements is output for each sub-region.

[0074] Step S32: The noisy convolution kernel is used to traverse the feature image to be detected layer by layer, and feature activation is performed on the common feature image region to be enhanced to obtain the feature image to be detected after abnormal noise enhancement.

[0075] It is understandable that the common feature image region to be enhanced refers to the part of the feature image region to be detected that has the same properties as the noise convolution kernel.

[0076] It should be noted that, in the specific implementation, the feature activation process for the common feature image regions to be enhanced involves using a layer-by-layer convolution operation. This allows convolutional kernels containing tampered noise information to activate the tampered region information in the image to be detected, thus enabling its representation during model training and inference. In other words, the feature image to be detected after anomalous noise enhancement is essentially a feature image obtained by enhancing the tampered region information.

[0077] This embodiment performs average pooling on the globally enhanced noise feature image to obtain a noise convolution kernel, and then uses it to traverse the feature image to be detected layer by layer to obtain the common feature image region information to be enhanced. Finally, the common feature activation is used to obtain the abnormal noise enhanced feature image to be detected, which improves the accuracy of subsequent image detection.

[0078] refer to Figure 4 , Figure 4 This is a flowchart illustrating a third embodiment of an image detection method based on noise enhancement according to the present invention.

[0079] Based on the first embodiment described above, in this embodiment, step S40 specifically includes the following steps:

[0080] Step S41: Perform initial image detection on the feature image to be detected after anomalous noise enhancement to obtain the first image detection result;

[0081] It should be noted that in the specific implementation, the initial image detection is achieved through the target screening module in the trained image tampering detection model. It is mainly based on the 14*14*1024 noise enhancement feature map given in the feature enhancement module for initial screening. This method reduces the false negative rate in the initial screening process and outputs a 14*14*2048 enhanced feature map, which is the first image detection result.

[0082] It should also be noted that, in the specific implementation, the image tampering detection model after training generally divides the image detection and screening process into two stages: a coarse classification stage and a fine classification stage. The initial image detection is the coarse classification stage of this screening process, which is used to distinguish between the foreground and background in the image.

[0083] Step S42: Perform depth image detection on the first image detection result to obtain the image detection and recognition result.

[0084] It should be noted that, in the specific implementation, depth image detection is achieved through the multi-sensor module in the trained image tampering detection model. The multi-sensor module is specifically composed of three information attention sub-modules: a global information attention sub-module, a local information attention sub-module, and a cross-information attention sub-module. Depth image detection is a fine classification stage in the image detection screening process. It is used to detect and identify tampered and non-tampered regions on image patches, and enhances noise features from different perceptual perspectives through different convolution methods.

[0085] It should also be noted that the final image detection and recognition result is obtained by integrating the image detection results generated by the three information attention submodules: the global information attention submodule, the local information attention submodule, and the cross information attention submodule.

[0086] This embodiment achieves the detection and recognition of the image to be detected by progressively detecting the feature image after abnormal noise enhancement.

[0087] refer to Figure 5 , Figure 5 This is a flowchart illustrating the fourth embodiment of an image detection method based on noise enhancement according to the present invention.

[0088] Based on the third embodiment described above, in this embodiment, step S42 specifically includes the following steps:

[0089] Step S421: Perform global information focus on the detection result of the first image to obtain the global information recognition result;

[0090] It should be noted that, in the specific implementation, global information focus first expands the enhanced feature map (i.e., the first image detection result) and the noisy feature image block into a one-dimensional vector and concatenates them together, then performs subsequent image tampering classification, and outputs the corresponding image tampering classification data, i.e., the global information recognition result.

[0091] Step S422: Perform local information focus on the detection result of the first image to obtain the local information recognition result;

[0092] It should be noted that, in the specific implementation, local information focus is achieved by using the convolutional neural network in the trained image detection model to detect the tampered parts in the first image detection result through the characteristics of local perception and weight sharing. The detection result obtained is the local information recognition result.

[0093] Step S423: Perform cross-information attention on the detection results of the first image to obtain the cross-information recognition results;

[0094] It should be noted that in the specific implementation, the cross-information focus (i.e. the acquisition of channel cross-information) needs to be based on the fully convolutional network of the trained image tampering detection model. During the operation, each patch in the enhanced feature map (i.e. the first image detection result) will perform cross-correlation operation with each patch in the noisy feature image block to obtain the location information of the image tampering area, i.e. the cross-information recognition result.

[0095] Step S424: Integrate the global information recognition results, local information recognition results, and cross-information recognition results to obtain the image detection and recognition results.

[0096] Understandably, in the specific implementation, the global information recognition results, local information recognition results, and cross-information recognition results are integrated. Specifically, this is achieved through the interaction and mutual supplementation of the three information attention methods. The result enriches the spatial information, thereby enabling the image tampering detection model to obtain information for accurately locating the image tampering location.

[0097] This embodiment uses three information focus methods to detect different types of tampering in the first image detection result one by one, and finally obtains the image detection and recognition result, which improves the accuracy of detecting tampered images.

[0098] Furthermore, the specific training process of the trained image tampering detection model includes: acquiring a training image dataset and performing data preprocessing on the training image data in the training image dataset to obtain a processed training image dataset; extracting data features from the processed training image data in the processed training image dataset to obtain an image feature dataset; classifying the training image feature dataset to obtain an image feature training set and an image feature test set; importing the image feature training set into the image tampering detection model to be trained for model training to obtain the image tampering detection model to be detected; importing the image feature test set into the image tampering detection model to be detected for model detection, and outputting the trained image tampering detection model when the model detection result is qualified.

[0099] Understandably, in the specific implementation, data feature extraction is performed on the processed training image data in the processed training image dataset to obtain an image feature dataset. The specific steps include: first, performing a convolution operation with a noise kernel on each positive and negative image instance in the processed training image dataset to obtain the corresponding image feature data; then, integrating the corresponding image feature data of each positive and negative image instance to obtain the image feature dataset. The purpose of performing a convolution operation with a noise kernel on each positive and negative image instance is to obtain the noise tampering features of the image.

[0100] It should be noted that, in the specific implementation, data preprocessing refers to first cutting each training image in the training image dataset into multiple corresponding tampered image blocks and untampered image blocks, then binding each tampered image block with multiple untampered image blocks to form positive and negative instances, and finally integrating all the positive and negative instances to obtain the processed training image dataset.

[0101] It should also be noted that a qualified model detection result specifically means that the model's accuracy in detecting tampered images, detection speed, and accuracy in locating the tampered image location meet preset standards. For example, if the preset standards are: image tamper detection accuracy of no less than 95%; detection speed of no less than 60 images per minute; and tamper location accuracy of no less than 90%, then all indicators of the image tamper detection model to be tested must meet the above preset standards in order to output the trained image tamper detection model normally.

[0102] Further, a training image dataset is acquired, and the training image data in the training image dataset is preprocessed to obtain a processed training image dataset. The specific steps include: acquiring the training image dataset, and splitting each training image data in the training image dataset into tampered images to obtain corresponding tampered part images and multiple untampered part images; for each training image data, binding each tampered part image and its corresponding untampered part image to obtain positive and negative image instances corresponding to each tampered part image; and integrating the obtained positive and negative image instances to obtain the processed training image dataset.

[0103] It should be noted that, in the specific implementation, the positive and negative instances of the image are mainly used to emphasize the discriminative differences between the noise features of the tampered area and the noise features of the untampered area.

[0104] Furthermore, embodiments of the present invention also propose a computer-readable storage medium storing a computer program, wherein the storage medium stores a noise-enhanced image detection program, and when the noise-enhanced image detection program is executed by a processor, it implements the steps of the noise-enhanced image detection method described above.

[0105] Since this storage medium adopts all the technical solutions of all the above embodiments, it has at least all the beneficial effects brought about by the technical solutions of the above embodiments, which will not be repeated here.

[0106] Reference Figure 5 , Figure 5 This is a structural block diagram of the first embodiment of the image detection device based on noise enhancement according to the present invention.

[0107] like Figure 5As shown, the image detection device based on noise enhancement proposed in this embodiment of the invention includes:

[0108] Preprocessing module 10: Inputs the image to be detected into the trained image tampering detection model and performs global noise enhancement processing on the image to be detected to obtain the processed image to be detected;

[0109] Feature extraction module 20: performs high-dimensional feature extraction on the processed image to be detected to obtain the feature image to be detected;

[0110] Noise enhancement module 30: Performs local noise enhancement processing on the feature image to be detected to obtain the feature image to be detected after abnormal noise enhancement;

[0111] Image detection module 40: Detects the feature image to be detected after abnormal noise enhancement, and obtains the image detection and recognition result.

[0112] This embodiment inputs the image to be detected into a trained image tampering detection model, and performs global noise enhancement on the image to obtain a processed image. High-dimensional feature extraction is then performed on the processed image to obtain a feature image to be detected. Local noise enhancement is then performed on the feature image to obtain a feature image with enhanced abnormal noise. Finally, the enhanced feature image is detected to obtain the image detection and recognition result. This embodiment, based on the trained image tampering detection model, first performs global noise enhancement on the image to be detected, thereby strengthening the noise features of the tampered portion of the image. Then, feature extraction and local noise enhancement are performed to obtain the feature image to be detected. Finally, detection is performed to obtain the corresponding detection result, improving the accuracy of tampering image detection.

[0113] It should be understood that the above are merely illustrative examples and do not constitute any limitation on the technical solutions of the present invention. In specific applications, those skilled in the art can make settings as needed, and the present invention does not impose any restrictions on this.

[0114] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this invention. In practical applications, those skilled in the art can select some or all of the workflow to achieve the purpose of this embodiment according to actual needs, and no restrictions are imposed here.

[0115] In addition, for technical details not described in detail in this embodiment, please refer to the image detection method based on noise enhancement provided in any embodiment of the present invention, which will not be repeated here.

[0116] Furthermore, it should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0117] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0118] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory (ROM) / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0119] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. An image detection method based on noise enhancement, characterized in that, include: The image to be detected is input into the trained image tampering detection model, and global noise enhancement is performed on the image to be detected to obtain the processed image to be detected. The trained image tampering detection model includes a Noise Conv, a multi-parameter convolutional layer, a feature enhancement module, and a detection module. The Noise Conv consists of an SRM kernel, a Bayar kernel, and a Command kernel, and is used to perform global noise enhancement on the image to be detected. The multi-parameter convolutional layer is used to extract high-dimensional features, and the parameters are determined through deep learning training. The feature enhancement module generates noisy convolution kernels and traverses and activates common feature regions; The detection module performs initial screening by the target screening module and in-depth detection by the multi-sensor module, and outputs the image tampering detection results and tampering location information. High-dimensional feature extraction is performed on the processed image to be detected to obtain the feature image to be detected; The method involves performing local noise enhancement processing on the feature image to be detected to obtain a feature image to be detected enhanced with abnormal noise. Specific steps include: randomly cutting the processed feature image to be detected to obtain multiple noise feature image blocks; performing average pooling processing on the noise feature image blocks to obtain the noise convolution kernel; and traversing the feature image to be detected layer by layer through the noise convolution kernel, and performing feature activation on the common feature image regions to be enhanced to obtain the feature image to be detected enhanced with abnormal noise. The process of detecting the enhanced image of the feature to be detected by the abnormal noise to obtain an image detection and recognition result includes: performing initial image detection on the enhanced image of the feature to be detected by the abnormal noise to obtain a first image detection result; Global information focus is applied to the detection result of the first image to obtain the global information recognition result; Local information focus is applied to the detection result of the first image to obtain the local information recognition result; Cross-information attention is applied to the first image detection result to obtain the cross-information recognition result; The global information recognition result, the local information recognition result, and the cross-information recognition result are integrated to obtain the image detection and recognition result.

2. The image detection method based on noise enhancement according to claim 1, characterized in that, The trained image tampering detection model includes the following specific training process: Obtain the training image dataset and perform data preprocessing on the training image data in the training image dataset to obtain the processed training image dataset; Data feature extraction is performed on the processed training image data in the processed training image dataset to obtain the image feature dataset; The training image feature dataset is classified to obtain an image feature training set and an image feature test set; The image feature training set is imported into the image tampering detection model to be trained for model training, thereby obtaining the image tampering detection model to be detected. The image feature test set is imported into the image tampering detection model to be detected for model detection, and the trained image tampering detection model is output when the model detection result is qualified.

3. The image detection method based on noise enhancement according to claim 2, characterized in that, Obtain the training image dataset, and perform data preprocessing on the training image data in the training image dataset to obtain the processed training image dataset. Specific steps include: Obtain the training image dataset, and split each training image data in the training image dataset into tampered images to obtain the corresponding tampered part image and multiple untampered part images; For each training image data, each tampered image and its corresponding untampered image are bound together to obtain the positive and negative image instances corresponding to each tampered image. The obtained positive and negative image instances are integrated to obtain the processed training image dataset.

4. An image detection device based on noise enhancement, characterized in that, The noise-enhanced image detection device includes: The preprocessing module inputs the image to be detected into the trained image tampering detection model and performs global noise enhancement on the image to be detected, resulting in a processed image. The trained image tampering detection model includes a Noise Conv, a multi-parameter convolutional layer, a feature enhancement module, and a detection module. The Noise Conv consists of an SRM kernel, a Bayar kernel, and a Command kernel, and is used to perform global noise enhancement on the image to be detected. The multi-parameter convolutional layer is used to extract high-dimensional features, and the parameters are determined through deep learning training. The feature enhancement module generates a noise convolutional kernel and traverses and activates common feature regions. The detection module performs initial screening by a target screening module and deep detection by a multi-sensor module, outputting the image tampering detection result and tampering location information. Feature extraction module: performs high-dimensional feature extraction on the processed image to be detected to obtain the feature image to be detected; Noise enhancement module: performs local noise enhancement processing on the feature image to be detected to obtain a feature image to be detected after abnormal noise enhancement; the noise enhancement module is used to randomly cut the processed feature image to be detected to obtain multiple noise feature image blocks; performs average pooling processing on the noise feature image blocks to obtain the noise convolution kernel; and traverses the feature image to be detected layer by layer through the noise convolution kernel, and performs feature activation on the common feature image regions to be enhanced to obtain the feature image to be detected after abnormal noise enhancement. Image detection module: Detects the enhanced image of the feature to be detected after abnormal noise, and obtains an image detection and recognition result; The image detection module is used to perform initial image detection on the enhanced image of the feature to be detected after abnormal noise, and obtain a first image detection result; Global information focus is applied to the detection result of the first image to obtain the global information recognition result; Local information focus is applied to the detection result of the first image to obtain the local information recognition result; Cross-information attention is applied to the first image detection result to obtain the cross-information recognition result; The global information recognition result, the local information recognition result, and the cross-information recognition result are integrated to obtain the image detection and recognition result.

5. An image detection device based on noise enhancement, characterized in that, The noise-enhanced image detection device includes: a memory, a processor, and a noise-enhanced image detection program stored in the memory and executable on the processor, wherein the noise-enhanced image detection program is configured to implement the noise-enhanced image detection method according to any one of claims 1 to 3.

6. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it is able to implement the steps of the noise-enhanced image detection method according to any one of claims 1 to 3.

Images