Image steganography detection method, device and equipment

Abstract

The application provides an image steganography detection method, device and equipment, and relates to the technical field of image processing. The image steganography detection method comprises the following steps: performing multi-dimensional feature extraction on a to-be-processed image to obtain image features in multiple dimensions; performing deep feature extraction on the image features in multiple dimensions to obtain deep features in multiple scales; fusing the deep features in multiple scales to obtain fused features; and performing classification prediction on the fused features to obtain a prediction result, which is used to represent whether the to-be-processed image is a steganography image. The application extracts multi-dimensional features to capture feature information brought by different steganography algorithms, fuses the features to strengthen the feature information of steganography in the feature matrix, and finally judges whether the image is steganography through a prediction structure. According to the method provided by the application, the to-be-processed image is detected only once, multiple steganography algorithms can be analyzed, the accuracy of image steganography detection is improved, and the computational complexity is effectively reduced.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more specifically, to an image steganalysis detection method, apparatus, and device. Background Technology

[0002] With the development of network technology, the importance of network security has become increasingly prominent across various industries. In the field of image information security, there exists a method of covert information transmission called image steganography. Image steganography allows information to be written into images in ways difficult for the human eye to observe, thus achieving covert information transmission. To effectively maintain network security, image steganography detection can be performed to determine whether an image contains steganographic information, thereby filtering out steganographic images with hidden information and preventing the transmission of non-compliant hidden information over the network.

[0003] Traditional techniques employ various dedicated steganalysis detection methods for different dimensions of steganography. Examples include Sample Pairs Analysis (SPA), an algorithm for Least Significant Bit (LSB) steganography in the image spatial domain; and color scrambling analysis methods for palette image steganography. Due to the diversity of steganography algorithms, algorithms targeting only one dimension have relatively low accuracy.

[0004] To ensure the accuracy of steganalysis detection for a single image, multiple steganalysis detection methods are required, and the image needs to be checked multiple times, which greatly increases the computational complexity of image steganalysis detection. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of the prior art by providing an image steganalysis detection method, apparatus, and device, so that multiple steganalysis methods can be detected in an image only once, thereby improving the accuracy of image steganalysis detection and greatly reducing the computational complexity of image steganalysis detection.

[0006] To achieve the above objectives, the technical solutions adopted in the embodiments of this application are as follows:

[0007] In a first aspect, embodiments of this application provide an image steganalysis detection method, including:

[0008] Multidimensional feature extraction is performed on the image to be processed to obtain image features in multiple dimensions;

[0009] Deep feature extraction is performed on the image features of the multiple dimensions to obtain deep features at multiple scales;

[0010] The depth features at the multiple scales are fused to obtain fused features;

[0011] The fused features are classified and predicted to obtain a prediction result, which is used to indicate whether the image to be processed is a steganalysis image.

[0012] In an optional implementation, before performing depth feature extraction on the multiple dimensions of image features to obtain depth features at multiple scales, the method further includes:

[0013] The image features from the multiple dimensions are concatenated to obtain a feature matrix;

[0014] The process of extracting deep features from the image features of the multiple dimensions yields deep features at multiple scales, including:

[0015] Deep feature extraction is performed on the feature matrix to obtain the deep features at multiple scales.

[0016] In an optional implementation, the step of performing multi-dimensional feature extraction on the image to be processed to obtain image features in multiple dimensions includes:

[0017] Based on the type of the image to be processed, a feature extraction method corresponding to the type is used to perform multi-dimensional feature extraction on the image to be processed, thereby obtaining the image features of the multiple dimensions;

[0018] The multi-dimensional image features include: color dimension features, frequency domain dimension features, and color index parity features.

[0019] In an optional implementation, the step of performing multi-dimensional feature extraction on the image to be processed according to the type of the image to be processed, using the feature extraction method corresponding to the type, to obtain the image features of the multiple dimensions, includes:

[0020] If the image to be processed is an uncompressed image or a lossless compressed image, then the color dimension features of the image to be processed are obtained;

[0021] The frequency domain dimension features are obtained based on the preset frequency domain dimension parameters;

[0022] The odd / even characteristics of the color index are obtained based on the preset color index parameters.

[0023] In an optional implementation, obtaining the color dimension features of the image to be processed includes:

[0024] If the image to be processed is a grayscale image, then the brightness features of the image to be processed are extracted, and the color dimension is assigned a value according to the brightness features to obtain the color dimension features.

[0025] Alternatively, if the image to be processed is a true-color image, then the color dimension features of the image to be processed are extracted.

[0026] In an optional implementation, the step of performing multi-dimensional feature extraction on the image to be processed according to the type of the image to be processed, using the feature extraction method corresponding to the type, to obtain the image features of the multiple dimensions, includes:

[0027] If the image to be processed is a color-compressed image, then extract the color dimension features of the image to be processed;

[0028] Based on the color dimension features, the color index parity features are obtained;

[0029] The frequency domain dimension features are obtained based on the preset frequency domain dimension parameters.

[0030] In an optional implementation, the step of performing multi-dimensional feature extraction on the image to be processed according to the type of the image to be processed, using the feature extraction method corresponding to the type, to obtain the image features of the multiple dimensions, includes:

[0031] If the image to be processed is a frequency domain compressed image, then a discrete cosine transform is performed on the image to be processed to obtain the frequency domain dimension features;

[0032] The color dimension features are obtained by parsing the frequency domain dimension features;

[0033] The odd / even characteristics of the color index are obtained based on the preset color index parameters.

[0034] In an optional implementation, fusing the depth features at the multiple scales to obtain fused features includes:

[0035] According to a preset scale, at least one sampling feature of the preset scale is sampled from depth features of at least one other scale; wherein, the at least one other scale is a scale other than the smallest scale among the plurality of scales;

[0036] The at least one sampled feature is concatenated to obtain a concatenated feature;

[0037] The spliced ​​features are subjected to dimensionality reduction convolution to obtain target features with the same number of channels as the depth features at the minimum scale;

[0038] Global max pooling is performed on the depth features at the minimum scale to obtain the channel coefficients;

[0039] The target feature and the channel coefficients are multiplied by a channel multiplication operation to obtain the fused feature.

[0040] Secondly, embodiments of this application also provide an image steganalysis detection device, comprising:

[0041] The first feature extraction module is used to extract multi-dimensional features from the image to be processed, thereby obtaining image features in multiple dimensions.

[0042] The second feature extraction module is used to extract deep features from the image features of the multiple dimensions to obtain deep features at multiple scales.

[0043] The feature fusion module is used to fuse the depth features at multiple scales to obtain fused features;

[0044] The prediction module is used to classify and predict the fused features to obtain a prediction result, which is used to indicate whether the image to be processed is a steganalysis image.

[0045] Thirdly, the present invention provides an electronic device, comprising: a processor, a storage medium, and a bus, wherein the storage medium stores program instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the program instructions to perform the steps of the image steganalysis detection method as described in any of the foregoing embodiments.

[0046] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the image steganalysis detection method as described in any of the foregoing embodiments.

[0047] The beneficial effects of this application are:

[0048] This application provides an image steganalysis detection method, apparatus, and device, comprising: extracting multi-dimensional features from an image to be processed to obtain multi-dimensional image features; extracting depth features from the multi-dimensional image features to obtain depth features at multiple scales; fusing the depth features at multiple scales to obtain fused features; and then classifying and predicting the fused features to obtain a prediction result, the prediction result being used to indicate whether the image to be processed is a steganalysis image. Since the method of this application extracts multi-dimensional features from the image to be processed, including multi-dimensional feature information from multiple steganalysis locations and methods, and further, by extracting and fusing depth features at multiple scales based on multi-dimensional image features, the response of steganalysis information features in the fused features is strengthened, while the response of normal features is suppressed and weakened. Therefore, classifying and detecting the fused features avoids false detections, eliminates the need for multiple steganalysis detections, and enables the detection of multiple steganalysis methods, thus improving the accuracy of image steganalysis detection and significantly reducing its computational complexity. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 A schematic flowchart of an image steganalysis detection method provided in an embodiment of this application;

[0051] Figure 2 A structural diagram of a backbone network for deep feature extraction provided in an embodiment of this application;

[0052] Figure 3 A flowchart illustrating another image steganalysis detection method provided in this application embodiment;

[0053] Figure 4 This is a schematic diagram of a multi-dimensional feature extraction process in an image steganalysis method provided in an embodiment of this application;

[0054] Figure 5 This is a schematic diagram illustrating another multidimensional feature extraction process in an image steganalysis method provided in this application embodiment;

[0055] Figure 6 This is a schematic diagram of another multi-dimensional feature extraction process in an image steganalysis detection method provided in this application embodiment;

[0056] Figure 7 A schematic flowchart illustrating another image steganalysis detection method provided in this application embodiment;

[0057] Figure 8 A structural diagram of feature fusion provided in an embodiment of this application;

[0058] Figure 9 This is a schematic diagram of the functional modules of an image steganalysis detection device provided in an embodiment of this application;

[0059] Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0061] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0062] In the description of this application, it should be noted that if the terms "upper", "lower", etc. appear to indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship that the product of this application is usually placed in, it is only for the convenience of describing this application and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application.

[0063] Furthermore, the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Additionally, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0064] It should be noted that, where there is no conflict, the features in the embodiments of this application can be combined with each other.

[0065] To improve the accuracy of image steganalysis detection and reduce computational complexity, this application provides an image steganalysis detection method. This method can perform detection simultaneously through multiple feature dimensions in a single steganalysis detection process, which ensures the accuracy of steganalysis detection and reduces the computational complexity of the detection process.

[0066] The image steganalysis method provided in this application, with reference to the accompanying drawings and specific examples, will be explained in detail below. The image steganalysis method provided in this application can be implemented by a computer device pre-installed with a preset image steganalysis algorithm or detection software, by running the algorithm or software. The computer device can be, for example, a server or a terminal, and the terminal can be a user computer. Figure 1 This is a flowchart illustrating an image steganalysis method provided in an embodiment of this application; as shown below. Figure 1 As shown, the method includes:

[0067] S101. Perform multi-dimensional feature extraction on the image to be processed to obtain image features in multiple dimensions.

[0068] In this embodiment, multi-dimensional image features can be obtained by performing multi-dimensional feature extraction on the image to be processed during the preprocessing stage. For example, multi-dimensional image features may include at least two of the following dimensions: color dimension features, frequency domain dimension features, and color index parity features. The color dimension features include features of multiple color channels, such as red (R) channel features, green (G) channel features, and blue (B) channel features. The frequency domain dimension features can also be called frequency domain dimension features, and can be represented as F. The color index parity features can be represented as P.

[0069] In this step, during the feature extraction process of the image to be processed, image features of multiple dimensions are extracted. Since each dimension of image features may contain hidden steganographic information, the hidden steganographic information can be extracted and displayed according to the different steganographic methods and the location of the hidden steganographic information in the steganographic dimension of the image to be processed. This yields image features of multiple dimensions, which facilitates subsequent image steganography detection.

[0070] S102. Perform deep feature extraction on image features of multiple dimensions to obtain deep features at multiple scales.

[0071] In possible implementation examples, a pre-trained backbone network can be used for deep feature extraction, such as a deep residual network (ResNet) or a lightweight convolutional network (MobileNet).

[0072] The following example, with reference to the accompanying diagram, illustrates the use of a backbone network for deep feature extraction. Figure 2 This is a structural diagram of a backbone network for deep feature extraction provided in an embodiment of this application. For example... Figure 2 As shown, the backbone network may include: convolutional block 1, N-1 convolutional blocks 2, and convolutional block 3. N is an integer greater than or equal to 2.

[0073] In this structure, N-1 convolutional blocks 2 are connected sequentially. The output of convolutional block 1 is connected to the input of the first convolutional block among the N-1 convolutional blocks 2, and the output of the last convolutional block among the N-1 convolutional blocks 2 is connected to the input of convolutional block 3. The input of convolutional block 1 is the input of the backbone network.

[0074] N-1 convolutional blocks are used to output the first-scale feature M1, the second-scale feature M2, ..., the N-1th-scale feature M1, respectively. N-1 And convolutional block 3 is used to output the Nth scale feature M.N Then the output feature D obtained using the backbone network can be expressed as D=[M1,M2,......,M n Among the N scale depth features, the first scale feature M1 is the largest scale depth feature, and the Nth scale feature M... N The depth feature is the smallest in size.

[0075] For example, to target random location steganography and adaptive steganography in a steganography algorithm, the output feature D may include: four depth features at different scales, i.e., N = 4, to capture depth steganography features of regions at different locations and of different sizes.

[0076] In the backbone network, the number of input and output channels of each convolutional block can be flexibly configured. The configuration principle is that the number of input channels should match the number of output channels of the previous convolutional block. In the backbone network, the number of input and / or output channels of the convolutional blocks can be increased in advance according to the complexity of the steganography algorithm and the number of steganography algorithms in the image to be detected, and vice versa.

[0077] In the aforementioned backbone network, convolutional block 1 can have, for example, 5 input channels and 32 output channels, with a size of 3×3 and a stride of 2. Each convolutional block 2 may sequentially include: a 3×3 convolutional layer, a first normalization layer, a first activation layer, a 1×1 convolutional layer, a second normalization layer, and a second activation layer, wherein the number of input and output channels of the convolutional layers can be flexibly configured. Convolutional block 3 has the same structure as the convolutional blocks, except that the size of the first convolutional layer is 5×5.

[0078] S103. Fuse depth features at multiple scales to obtain fused features.

[0079] In this embodiment, by fusing depth features at multiple scales to obtain fused features, the response of stegographic information features can be strengthened, while the response of normal features can be suppressed and weakened, avoiding false detection and helping to improve the detection accuracy of classification detection.

[0080] S104. Classify and predict the fused features to obtain the prediction results, which are used to indicate whether the image to be processed is a steganalysis image.

[0081] For example, a prediction structure can be used to classify fused features and obtain the prediction result. This prediction structure may, for example, include: a global pooling layer, two fully connected layers, and a normalization layer. The output dimension of the second fully connected layer can be pre-set to the image category. In one example, if the image category can be divided into two classes: images that are steganized and images that are not steganized, then its output dimension is set to 2.

[0082] Correspondingly, the output of the second fully connected layer can include the scores of the two image categories, and the image category with the highest score is the final prediction result.

[0083] In summary, this application provides an image steganalysis detection method. It extracts multi-dimensional features from the image to be processed, obtaining multi-dimensional image features. Then, it extracts depth features from these multi-dimensional image features, obtaining depth features at multiple scales. These multi-scale depth features are then fused to obtain fused features. Finally, the fused features are classified and predicted to obtain a prediction result, which indicates whether the image to be processed is a steganalysis image. Because this method extracts multi-dimensional features from the image to be processed, including multi-dimensional feature information from various steganalysis locations and methods, and further, by extracting and fusing depth features at multiple scales based on multi-dimensional image features, the response of steganalysis information features in the fused features is strengthened, while the response of normal features is suppressed and weakened. Therefore, classification and detection of the fused features can avoid false detections. Multiple steganalysis detections can be performed without multiple steganalysis detections, thus improving the accuracy of image steganalysis detection and significantly reducing its computational complexity.

[0084] Based on the image steganalysis detection method provided in the above embodiments, this application embodiment also provides a possible implementation of the image steganalysis detection method by providing an implementation example of depth features. Figure 3 This is a schematic flowchart illustrating another image steganalysis method provided in an embodiment of this application. Optionally, as... Figure 3 As shown, before performing deep feature extraction on multi-dimensional image features in step S102 of the above method to obtain multi-scale deep features, the method further includes:

[0085] S301. Concatenate the image features of multiple dimensions to obtain the feature matrix.

[0086] To facilitate deep feature extraction, image features from multiple dimensions need to be concatenated to obtain a feature matrix. As shown above, image features from multiple dimensions may include at least two of the following: color dimension features, frequency domain dimension features, and color index parity features.

[0087] Assuming the image features across multiple dimensions include color dimension features, frequency domain dimension features, and color index parity features, then each pixel has 5 feature channels. Concatenating these multiple dimension image features yields a 5×H×W dimension feature matrix V: V = [R, G, B, F, P]. Here, H and W represent the image height and width, and 5 represents the number of feature channels per pixel.

[0088] S302. Perform deep feature extraction on the feature matrix to obtain deep features at multiple scales.

[0089] When a feature matrix V is formed by stitching together image features from multiple dimensions, it can be directly input into a feature extraction network for deep feature extraction.

[0090] In the image steganalysis shown in any of the above embodiments, for example, based on the type of the image to be processed, a feature extraction method corresponding to the type can be used to extract multi-dimensional features of the image to be processed, thereby obtaining image features of multiple dimensions. For example, the type of the image to be processed may include: uncompressed image or lossless compressed image, color compressed image, and frequency domain compressed image.

[0091] The following is in conjunction with the appendix Figure 4 Appendix Figure 5 Appendix Figure 6 Examples are provided to illustrate the feature extraction methods for different types of images to be processed. Figure 4 This is a schematic diagram illustrating a multi-dimensional feature extraction process in an image steganalysis method provided in an embodiment of this application. For example... Figure 4 As shown, based on the type of the image to be processed, a feature extraction method corresponding to the type is used to perform multi-dimensional feature extraction on the image to be processed, resulting in image features of multiple dimensions, including:

[0092] S401. If the image to be processed is an uncompressed image or a lossless compressed image, then obtain the color dimension features of the image to be processed.

[0093] The image to be processed can be an uncompressed image or a lossless compressed image, specifically categorized as grayscale or true-color images. Storage formats for uncompressed or lossless compressed images include, for example, png, tiff, bmp, raw, and svg.

[0094] In one possible implementation, if the image to be processed is a grayscale image, the brightness features of the image to be processed are extracted, and the color dimension is assigned a value based on the brightness features to obtain the color dimension features.

[0095] Since a grayscale image has only a single-channel brightness feature, it can be represented as Y. According to the general principle that a single-channel image can be converted into a three-channel image, the single-channel brightness feature Y can be assigned to its color dimension feature, which can be represented as: R = G = B = Y.

[0096] In another possible implementation, if the image to be processed is a true-color image, then the color dimension features of the image to be processed are extracted.

[0097] Since true-color images are three-channel images with color dimension features, their color dimension features can be directly extracted.

[0098] S402. Obtain the frequency domain dimension features based on the preset frequency domain dimension parameters.

[0099] S403. Obtain the odd / even characteristics of the color index based on the preset color index parameters.

[0100] The preset frequency domain dimension parameter and preset color index parameter can be, for example, 0. Accordingly, the corresponding frequency domain dimension feature and color index parity feature are set to the default 0 matrix.

[0101] In the image steganalysis detection method provided in this application embodiment, for uncompressed or lossless compressed images, only color dimension features need to be obtained, while for frequency domain dimension features and color index parity features, they can be directly configured as preset parameters, reducing their computational complexity.

[0102] Figure 5 This is a schematic diagram illustrating another multi-dimensional feature extraction process in an image steganalysis method provided in this application embodiment. For example... Figure 5 As shown, based on the type of the image to be processed, a feature extraction method corresponding to the type is used to perform multi-dimensional feature extraction on the image to be processed, resulting in image features of multiple dimensions, including:

[0103] S501. If the image to be processed is a color compressed image, then extract the color dimension features of the image to be processed.

[0104] The image to be processed can be a color-compressed image, such as a palette image. Typically, the number of colors in the image is compressed to within 256 colors, and a color table is used to define each color in the color-compressed image; this can also be called a false-color image. Then, by matching the color index values ​​with the corresponding color table, the color dimension features of the color-compressed image are obtained. The storage format of a color-compressed image can be, for example, GIF.

[0105] S502. Based on the color dimension features, obtain the color index parity features.

[0106] The color dimension features obtained above through color index values ​​and corresponding color tables are then used to calculate a binary image as the color index parity feature based on the parity of the sum of color values ​​in the color dimension features. This can be represented as P:

[0107]

[0108] In the above formula, P(x,y) represents the feature value of the x-th row and y-th column after parity calculation, R(x,y) represents the red color value of the pixel in the x-th row and y-th column, G(x,y) represents the green color value of the pixel in the x-th row and y-th column, and B(x,y) represents the blue color value of the pixel in the x-th row and y-th column.

[0109] S503. Obtain the frequency domain dimension features based on the preset frequency domain dimension parameters.

[0110] The preset frequency domain dimension parameter can be, for example, 0, and correspondingly, the corresponding frequency domain dimension feature is set to the default 0 matrix.

[0111] In the image steganalysis detection method provided in this application embodiment, when performing multi-dimensional feature extraction on a color-compressed image, the color dimension features are extracted, and a binary image is obtained by calculating the parity of the sum of color values ​​in the color dimension features. This yields the color index parity features of the color-compressed image. For the frequency domain dimension features, they can be directly configured with preset parameters, reducing their computational complexity.

[0112] Figure 6 This is a schematic diagram illustrating another multi-dimensional feature extraction process in an image steganalysis method provided in this application embodiment. For example... Figure 6 As shown, based on the type of the image to be processed, a feature extraction method corresponding to the type is used to perform multi-dimensional feature extraction on the image to be processed, resulting in image features of multiple dimensions, including:

[0113] S601. If the image to be processed is a frequency domain compressed image, then perform a discrete cosine transform on the image to be processed to obtain the frequency domain dimension features.

[0114] The image to be processed can be a frequency domain compressed image, typically stored in JPG or JPEG format. Frequency domain compression involves converting a spatial image to a frequency domain image, then performing lossy quantization using a quantization matrix, and finally applying a discrete cosine transform to the compressed image to obtain its frequency domain dimensional features.

[0115] S602. Analyze the frequency domain dimension features to obtain the color dimension features.

[0116] The obtained frequency domain dimension features are analyzed, and the analyzed colors are used as the color dimension features of the frequency domain compressed image, thus obtaining the color dimension features of the frequency domain compressed image.

[0117] S603. Obtain the odd / even characteristics of the color index based on the preset color index parameters.

[0118] The preset color index parameter can be 0, for example, and correspondingly, the color index parity feature is set to the default 0 matrix.

[0119] In the image steganalysis detection method provided in this application embodiment, when performing multi-dimensional feature extraction on a frequency domain compressed image, only the frequency domain dimension features are extracted, and then the color dimension features are obtained by parsing the frequency domain dimension features. As for the color index parity feature, it can be directly configured as a preset parameter, thereby reducing its computational complexity.

[0120] The feature extraction methods provided in this application for different types of images to be processed include three different types of images to be processed: uncompressed or lossless compressed images, color compressed images, and frequency domain compressed images. These methods extract image features of multiple dimensions of the corresponding images, capturing image information from multiple channels and locations where steganographic information may exist, as well as image information from multiple steganographic methods. This enables the detection of multiple steganographic methods, improves the accuracy of image steganography detection, and reduces the computational complexity in the multidimensional feature extraction process.

[0121] Based on the image steganalysis detection method provided in any of the above embodiments, this application also provides an implementation method for feature fusion of depth features at multiple scales. Figure 7 This is a schematic flowchart illustrating another image steganalysis detection method provided in an embodiment of this application. Figure 7 As shown, depth features at multiple scales are fused to obtain fused features, including:

[0122] S701. Based on a preset scale, sample at least one sampling feature of the preset scale from depth features of at least one other scale; wherein, at least one other scale is a scale other than the smallest scale among multiple scales.

[0123] The feature matrix obtained by stitching together image features from multiple dimensions is input into the backbone network for deep feature extraction, resulting in deep features at multiple scales.

[0124] This embodiment is illustrated with a structural diagram of feature fusion. Figure 8 This is a structural diagram of feature fusion provided in an embodiment of this application. For example, such as... Figure 8 As shown, depth features at multiple scales may include, for example, a first scale feature M1, a second scale feature M2, a third scale feature M3, and a fourth scale feature M4, wherein the first scale feature M1 is the depth feature at the largest scale, and the fourth scale feature M4 is the depth feature at the smallest size.

[0125] Based on a preset scale, the obtained first-scale feature M1, second-scale feature M2, and third-scale feature M3 are subjected to interpolation sampling to obtain first-sampled feature M1', second-sampled feature M2', and third-sampled feature M3' of the same size as the preset size. The preset size can be, for example, the scale of the third-scale feature M3.

[0126] S702. At least one sampled feature is concatenated to obtain a concatenated feature.

[0127] Continue to refer to Figure 8 The first sampling feature M1', the second sampling feature M2', and the third sampling feature M3' are horizontally concatenated to obtain the concatenated feature M5.

[0128] S703. Perform dimensionality reduction convolution on the spliced ​​features to obtain target features with the same number of channels as the depth features at the minimum scale.

[0129] Continue to refer to Figure 8 Dimensionality reduction convolution is performed on the spliced ​​feature M5 to obtain the target feature M6 with the same number of channels as the minimum scale depth feature, that is, the fourth scale feature M4.

[0130] S704. Perform global max pooling on the depth features at the smallest scale to obtain the channel coefficients.

[0131] Global max pooling is performed on the minimum-scale depth feature, i.e., the fourth-scale feature M4, to obtain the channel coefficients M4'.

[0132] S705. Perform channel multiplication on the target features and channel coefficients to obtain the fused features.

[0133] Finally, the target feature M6 obtained through step S703 and the channel coefficient M4' obtained through step S704 are multiplied by channel to obtain the fused feature M7.

[0134] The deep feature fusion method provided in this application extracts deep features at multiple scales as input, performs difference sampling on the deep features at multiple scales, and then horizontally concatenates them to obtain concatenated features. Dimensionality reduction convolution is then performed to obtain a target matrix with the same number of channels as the smallest-sized deep feature. This target matrix is ​​then multiplied by the channel response coefficients obtained from global max pooling of the smallest-sized deep feature, ultimately yielding the fused feature. This method enhances the response of steganalytic information features in the fused feature while suppressing and weakening the response of normal features. Therefore, it helps in the classification and detection of the fused feature and avoids false detections.

[0135] The following will continue to explain the apparatus, device and storage medium for image steganalysis provided in any of the above embodiments of this application. The specific implementation process and the resulting technical effects are the same as those in the corresponding method embodiments. For the sake of brevity, parts not mentioned in this embodiment can be referred to the corresponding content in the method embodiments.

[0136] Figure 9 This is a functional module diagram of an image steganalysis detection device provided in an embodiment of this application; as shown below. Figure 9 As shown, the image steganalysis detection device 100 includes:

[0137] The first feature extraction module 110 is used to perform multi-dimensional feature extraction on the image to be processed, and obtain image features in multiple dimensions.

[0138] The second feature extraction module 120 is used to extract deep features from image features of multiple dimensions to obtain deep features at multiple scales.

[0139] The feature fusion module 130 is used to fuse depth features at multiple scales to obtain fused features;

[0140] The prediction module 140 is used to classify and predict the fused features to obtain the prediction result, which is used to indicate whether the image to be processed is a steganalysis image.

[0141] In an optional embodiment, the image steganalysis detection device 100 further includes:

[0142] The stitching module is used to stitch together image features of multiple dimensions to obtain a feature matrix; and to extract deep features from image features of multiple dimensions to obtain deep features at multiple scales, including: extracting deep features from the feature matrix to obtain deep features at multiple scales.

[0143] In an optional implementation, the first feature extraction module 110 is specifically used to perform multi-dimensional feature extraction on the image to be processed by adopting a feature extraction method corresponding to the type of the image to be processed, thereby obtaining image features of multiple dimensions.

[0144] Multidimensional image features include: color dimension features, frequency domain dimension features, and color index parity features.

[0145] In an optional implementation, the first feature extraction module 110 is specifically used to obtain the color dimension features of the image to be processed if the image to be processed is an uncompressed image or a lossless compressed image; obtain the frequency domain dimension features according to the preset frequency domain dimension parameters; and obtain the color index parity features according to the preset color index parameters.

[0146] In an optional implementation, if the image to be processed is a grayscale image, the brightness features of the image to be processed are extracted, and the color dimension is assigned a value according to the brightness features to obtain the color dimension features; or, if the image to be processed is a true color image, the color dimension features of the image to be processed are extracted.

[0147] In an optional implementation, the first feature extraction module 110 is specifically used to extract the color dimension features of the image to be processed if the image to be processed is a color compressed image; obtain the color index parity features based on the color dimension features; and obtain the frequency domain dimension features based on the preset frequency domain dimension parameters.

[0148] In an optional implementation, the first feature extraction module 110 is further configured to perform discrete cosine transform on the image to be processed if the image to be processed is a frequency domain compressed image, to obtain frequency domain dimension features; to parse the frequency domain dimension features to obtain color dimension features; and to obtain color index parity features according to preset color index parameters.

[0149] In an optional implementation, the feature fusion module 130 is specifically configured to sample at least one sampled feature at a preset scale from depth features at at least one other scale, wherein the at least one other scale is a scale other than the smallest scale among multiple scales; concatenate the at least one sampled feature to obtain a concatenated feature; perform dimensionality reduction convolution on the concatenated feature to obtain a target feature with the same number of channels as the depth feature at the smallest scale; perform global max pooling on the depth feature at the smallest scale to obtain channel coefficients; and perform channel multiplication on the target feature and the channel coefficients to obtain a fused feature.

[0150] The above-described device is used to execute the method provided in the foregoing embodiments, and its implementation principle and technical effect are similar, so they will not be described again here.

[0151] These modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more microprocessors, or one or more Field Programmable Gate Arrays (FPGAs). Alternatively, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together as a system-on-a-chip (SOC).

[0152] Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. This electronic device can be used for image steganalysis detection. Figure 10 As shown, the electronic device includes: a processor 210, a storage medium 220, and a bus 230.

[0153] Storage medium 220 stores machine-readable instructions executable by processor 210. When the electronic device is running, processor 210 communicates with storage medium 220 via bus 230, and processor 210 executes the machine-readable instructions to perform the steps of the above method embodiment. The specific implementation and technical effects are similar and will not be described again here.

[0154] Optionally, this application also provides a storage medium 220, on which a computer program is stored. When the computer program is run by a processor, it executes the steps of the above-described method embodiments. The specific implementation and technical effects are similar, and will not be repeated here.

[0155] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0156] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0157] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0158] The integrated units implemented as software functional units described above can be stored in a computer-readable storage medium. These software functional units, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0159] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An image steganalysis detection method, characterized in that, include: Multidimensional feature extraction is performed on the image to be processed to obtain image features in multiple dimensions; Deep feature extraction is performed on the image features of the multiple dimensions to obtain deep features at multiple scales; The depth features at the multiple scales are fused to obtain fused features; The fused features are classified and predicted to obtain a prediction result, which is used to indicate whether the image to be processed is a stegana image. The process of fusing depth features at multiple scales to obtain fused features includes: According to a preset scale, at least one sampling feature of the preset scale is sampled from depth features of at least one other scale; wherein, the at least one other scale is a scale other than the smallest scale among the plurality of scales; The at least one sampled feature is concatenated to obtain a concatenated feature; The spliced ​​features are subjected to dimensionality reduction convolution to obtain target features with the same number of channels as the depth features at the minimum scale; Global max pooling is performed on the depth features at the minimum scale to obtain the channel coefficients; The target feature and the channel coefficients are multiplied by a channel multiplication operation to obtain the fused feature.

2. The method as described in claim 1, characterized in that, Before performing deep feature extraction on the image features of the multiple dimensions to obtain deep features at multiple scales, the method further includes: The image features from the multiple dimensions are concatenated to obtain a feature matrix; The process of extracting deep features from the image features of the multiple dimensions yields deep features at multiple scales, including: Deep feature extraction is performed on the feature matrix to obtain the deep features at multiple scales.

3. The method as described in claim 1, characterized in that, The image to be processed is subjected to multi-dimensional feature extraction to obtain image features in multiple dimensions, including: Based on the type of the image to be processed, a feature extraction method corresponding to the type is used to perform multi-dimensional feature extraction on the image to be processed, thereby obtaining the image features of the multiple dimensions; The multi-dimensional image features include: color dimension features, frequency domain dimension features, and color index parity features.

4. The method as described in claim 3, characterized in that, The step involves extracting multi-dimensional features from the image to be processed using a feature extraction method corresponding to the type of the image, thereby obtaining the multi-dimensional image features, including: If the image to be processed is an uncompressed image or a lossless compressed image, then the color dimension features of the image to be processed are obtained; The frequency domain dimension features are obtained based on the preset frequency domain dimension parameters; The odd / even characteristics of the color index are obtained based on the preset color index parameters.

5. The method as described in claim 4, characterized in that, The step of obtaining the color dimension features of the image to be processed includes: If the image to be processed is a grayscale image, then the brightness features of the image to be processed are extracted, and the color dimension is assigned a value according to the brightness features to obtain the color dimension features. Alternatively, if the image to be processed is a true-color image, then the color dimension features of the image to be processed are extracted.

6. The method as described in claim 3, characterized in that, The step involves extracting multi-dimensional features from the image to be processed using a feature extraction method corresponding to the type of the image, thereby obtaining the multi-dimensional image features, including: If the image to be processed is a color-compressed image, then extract the color dimension features of the image to be processed; Based on the color dimension features, the color index parity features are obtained; The frequency domain dimension features are obtained based on the preset frequency domain dimension parameters.

7. The method as described in claim 3, characterized in that, The step involves extracting multi-dimensional features from the image to be processed using a feature extraction method corresponding to the type of the image, thereby obtaining the multi-dimensional image features, including: If the image to be processed is a frequency domain compressed image, then a discrete cosine transform is performed on the image to be processed to obtain the frequency domain dimension features; The color dimension features are obtained by parsing the frequency domain dimension features; The odd / even characteristics of the color index are obtained based on the preset color index parameters.

8. An image steganalysis detection device, characterized in that, include: The first feature extraction module is used to extract multi-dimensional features from the image to be processed, thereby obtaining image features in multiple dimensions. The second feature extraction module is used to extract deep features from the image features of the multiple dimensions to obtain deep features at multiple scales. The feature fusion module is used to fuse the depth features at multiple scales to obtain fused features; The prediction module is used to classify and predict the fused features to obtain a prediction result, which is used to indicate whether the image to be processed is a steganalysis image. The feature fusion module is further configured to sample at least one sampling feature of the preset scale from depth features of at least one other scale according to a preset scale; wherein the at least one other scale is a scale other than the smallest scale among the plurality of scales; to concatenate the at least one sampling feature to obtain a concatenated feature; and to perform dimensionality reduction convolution on the concatenated feature to obtain a target feature with the same number of channels as the depth feature of the smallest scale. Global max pooling is performed on the depth features at the minimum scale to obtain channel coefficients; channel multiplication is performed on the target features and the channel coefficients to obtain the fused features.

9. An electronic device, characterized in that, include: The device includes a processor, a storage medium, and a bus. The storage medium stores program instructions executable by the processor. When the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the program instructions to perform the steps of the image steganalysis detection method as described in any one of claims 1 to 7.

Images