CN113763405A - An image detection method and device - Google Patents

Abstract

The invention discloses an image detection method and device, and relates to the technical field of computers. A specific implementation of the method includes: acquiring a training sample; wherein, the training sample includes: a training image, a region label and a boundary label; inputting the training image into a detection model to obtain a region detection result and a boundary detection result; The region label, the boundary label, the region detection result and the boundary detection result are used to train the detection model; based on the trained detection model, it is determined whether the detection image has been tampered with. This embodiment can improve detection accuracy.

Description

An image detection method and device

technical field

The present invention relates to the field of computer technology, and in particular, to an image detection method and device.

Background technique

In practical application scenarios, criminals synthesize the contents of multiple images into one image, which changes the original meaning of the image and misleads users. For example, in e-commerce platforms, merchants tamper with original images to attract consumers. Therefore, how to detect whether an image has been tampered with has become an urgent problem to be solved.

In the prior art, edge detection is used to identify whether an image has been tampered with.

However, this method only focuses on the local features of the image, and its detection accuracy is low.

SUMMARY OF THE INVENTION

In view of this, embodiments of the present invention provide an image detection method and apparatus, which can improve detection accuracy.

In a first aspect, an embodiment of the present invention provides an image detection method, including:

Obtain training samples; wherein, the training samples include: training images, region labels and boundary labels;

Inputting the training image into a detection model to obtain a region detection result and a boundary detection result;

train the detection model according to the region label, the boundary label, the region detection result and the boundary detection result;

Based on the trained detection model, it is determined whether the detection image has been tampered with.

Optionally,

The detection model includes: a feature extraction layer, a region detection layer and a boundary detection layer;

The training image is input into the detection model to obtain the region detection result and the boundary detection result, including:

inputting the training image into the feature extraction layer to extract a high-order feature map and a low-order feature map from the training image;

Inputting the high-order feature map and the low-order feature map into the region detection layer to obtain the region detection result;

The high-order feature map and the low-order feature map are input into the boundary detection layer to obtain the boundary detection result.

Optionally,

Inputting the training image into the feature extraction layer to extract high-order feature maps and low-order feature maps from the training images, including:

Inputting the training image into the backbone network to obtain the low-level feature map and the first feature map;

Extracting multi-scale features from the first feature map based on the multi-scale network to obtain multiple second feature maps;

After splicing the plurality of second feature maps, they are input into the first convolution layer to obtain the high-order feature maps;

The backbone network includes: a first multi-channel convolutional layer and a depthwise separable convolutional layer; the first convolutional layer is a 1×1 convolutional layer.

Optionally,

The multi-scale network includes: an atrous convolutional layer, a second convolutional layer and a pooling layer;

Wherein, the second convolutional layer is a 1×1 convolutional layer.

Optionally,

The region detection layer includes: a first feature fusion layer, a region anomaly analysis layer and a first result output layer;

Inputting the high-order feature map and the low-order feature map into the region detection layer to obtain the region detection result, including:

Inputting the high-order feature map and the low-order feature map into the first feature fusion layer to obtain a third feature map;

According to the third feature map and the regional anomaly analysis layer, a fourth feature map is determined; wherein, the fourth feature map is used to represent the pixel value difference between the tampered area and the background area in the third feature map;

The fourth feature map is input into the first result output layer to obtain the region detection result.

Optionally,

Inputting the high-order feature map and the low-order feature map into the first feature fusion layer to obtain a third feature map, including:

Input the low-level feature map into the third convolutional layer to obtain the fifth feature map;

Upsampling the high-order feature map to obtain a sixth feature map;

After splicing the fifth feature map and the sixth feature map, input the second multi-channel convolution layer to obtain the third feature map;

Wherein, the third convolutional layer is a 1×1 convolutional layer.

Optionally,

The determining a fourth feature map according to the third feature map and the regional anomaly analysis layer includes:

Calculate the average pixel value of the third feature map according to the pixel value of each pixel coordinate in the third feature map;

determining the difference between the pixel value of each of the pixel coordinates and the average pixel value;

Calculate the pixel value standard deviation of the third feature map according to the difference between the pixel value of each of the pixel coordinates and the average pixel value;

Calculate the normalized pixel value of each of the pixel coordinates according to the standard deviation of the pixel value and the difference between the pixel value of each of the pixel coordinates and the average pixel value;

The fourth feature map is determined according to the normalized pixel value of each of the pixel coordinates.

Optionally,

The inputting the fourth feature map into the first result output layer to obtain the region detection result includes:

Input the fourth feature map into the fourth convolution layer to obtain the seventh feature map;

Upsampling the seventh feature map to obtain an eighth feature map;

Inputting the eighth feature map into an activation function to obtain the region detection result;

Wherein, the fourth convolutional layer is a 1×1 convolutional layer.

Optionally,

The boundary detection layer includes: a second feature fusion layer, a boundary anomaly analysis layer and a second result output layer;

Inputting the high-order feature map and the low-order feature map into the boundary detection layer to obtain the boundary detection result, including:

Inputting the high-order feature map and the low-order feature map into the second feature fusion layer to obtain a ninth feature map;

According to the ninth feature map and the boundary anomaly analysis layer, determine a tenth feature map; wherein, the tenth feature map is used to represent the pixel value difference between the tampered area and the background area in the detection window;

The tenth feature map is input into the second result output layer to obtain the boundary detection result.

Optionally,

Inputting the high-order feature map and the low-order feature map into the second feature fusion layer to obtain a ninth feature map, including:

Input the low-level feature map into the fifth convolutional layer to obtain the eleventh feature map;

Upsampling the high-order feature map to obtain a twelfth feature map;

After splicing the eleventh feature map and the twelfth feature map, input the third multi-channel convolutional layer to obtain the ninth feature map;

Wherein, the fifth convolutional layer is a 1×1 convolutional layer.

Optionally,

The determination of the tenth feature map according to the ninth feature map and the boundary anomaly analysis layer includes:

Calculate the average pixel value of the detection window according to the pixel value of each pixel coordinate in the ninth feature map in the detection window;

determining the difference between the pixel value of each of the pixel coordinates and the average pixel value of the detection window where the pixel coordinates are located;

calculating the pixel value standard deviation of the ninth feature map;

According to the standard deviation of the pixel value, the difference between the pixel value of each of the pixel coordinates and the average pixel value of the detection window where the pixel coordinates are located, calculate the normalized pixel value of the pixel coordinates in the detection window;

The tenth feature map is determined according to the normalized pixel values of the pixel coordinates within the detection window.

Optionally,

The inputting the tenth feature map into the second result output layer to obtain the boundary detection result includes:

Input the tenth feature map into the sixth convolution layer to obtain the thirteenth feature map;

Upsampling the thirteenth feature map to obtain a fourteenth feature map;

Inputting the fourteenth feature map into an activation function to obtain the region detection result;

Wherein, the sixth convolutional layer is a 1×1 convolutional layer.

Optionally,

The acquiring training samples includes:

obtain the training image and the region label;

performing an expansion operation on the region label to obtain an expanded image;

performing an erosion operation on the region label to obtain an erosion image;

The boundary labels are determined from the dilation image and the erosion image.

Optionally,

Further includes:

Get pre-training samples;

pre-training the detection model based on the pre-training samples;

The described training image is input into the detection model to obtain the region detection result and the boundary detection result, including:

The training image is input into the pre-trained detection model to obtain the region detection result and the boundary detection result.

In a second aspect, an embodiment of the present invention provides an image detection device, including:

an acquisition module, configured to acquire training samples; wherein, the training samples include: training images, region labels and boundary labels;

a training module, configured to input the training image into a detection model to obtain a region detection result and a boundary detection result; train the detection according to the region label, the boundary label, the region detection result and the boundary detection result Model;

The detection module is configured to determine whether the detection image has been tampered with based on the trained detection model.

In a third aspect, an embodiment of the present invention provides an electronic device, including:

one or more processors;

storage means for storing one or more programs,

The one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method as described in any of the above embodiments.

In a fourth aspect, an embodiment of the present invention provides a computer-readable medium on which a computer program is stored, and when the program is executed by a processor, implements the method described in any of the foregoing embodiments.

One embodiment of the above invention has the following advantages or beneficial effects: boundary detection and area detection are performed on the image based on the detection model, and the area detection is based on the characteristic difference between the tampered area and the background area of the entire image, and the tampered area is identified. Overall features; Boundary detection identifies tampered boundaries based on feature differences on both sides of the tampered boundary, focusing on local features. Boundary detection can assist region detection, determine the tampered region more accurately, and improve the accuracy of image detection.

Further effects of the above non-conventional alternatives will be described below in conjunction with specific embodiments.

Description of drawings

The accompanying drawings are used for better understanding of the present invention and do not constitute an improper limitation of the present invention. in:

1 is a flowchart of an image detection method provided by an embodiment of the present invention;

2 is a flowchart of an image detection method provided by an embodiment of the present invention;

3 is an architecture diagram of a detection model provided by an embodiment of the present invention;

Figure 4 (a) is a schematic diagram of a region label provided by an embodiment of the present invention;

Figure 4(b) is a schematic diagram of a dilated image provided by an embodiment of the present invention;

Figure 4(c) is a schematic diagram of a corrosion image provided by an embodiment of the present invention;

4(d) is a schematic diagram of a border label provided by an embodiment of the present invention;

5 is a schematic structural diagram of a backbone network provided by an embodiment of the present invention;

6 is a schematic structural diagram of a multi-scale network provided by an embodiment of the present invention;

7 is a schematic structural diagram of an image detection apparatus provided by an embodiment of the present invention;

FIG. 8 is an exemplary system architecture diagram to which an embodiment of the present invention may be applied;

FIG. 9 is a schematic structural diagram of a computer system suitable for implementing a terminal device or a server according to an embodiment of the present invention.

Detailed ways

Exemplary embodiments of the present invention are described below with reference to the accompanying drawings, which include various details of the embodiments of the present invention to facilitate understanding and should be considered as exemplary only. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the invention. Also, descriptions of well-known functions and constructions are omitted from the following description for clarity and conciseness.

Edge detection focuses on detecting the local features of the image in the frame, and does not consider the feature difference between the tampered area and the background area of the entire image. Therefore, the accuracy of its detection results needs to be further improved.

In view of this, as shown in FIG. 1 , an embodiment of the present invention provides an image detection method, including:

Step 101: Obtain training samples; wherein, the training samples include: training images, region labels and boundary labels.

In order to improve the training effect, the training samples of the embodiment of the present invention come from the tampered image data set CASIA 2.0. CASIA 2.0 can provide more than 5,000 tampered images, and involves a variety of tampering methods and image formats, which can satisfy the training of the embodiment of the present invention. need. In practical application scenarios, the CASIA 1.0 data set with a relatively small number of samples can also be selected according to the actual situation.

The dataset includes training images and region labels, and boundary labels are determined from the region labels.

Step 102: Input the training image into the detection model to obtain the region detection result and the boundary detection result.

The area detection result is the predicted tampered area, and the boundary detection result is the predicted tampered boundary.

Step 103: Train the detection model according to the region label, the boundary label, the region detection result and the boundary detection result.

The loss value is determined according to the region label, the boundary label, the region detection result, the boundary detection result and the preset loss function; according to the loss value, the parameters of the detection model are adjusted.

In practical application scenarios, in order to ensure the prediction quality of the detection model, during the training process, the detection model is tested with test samples to determine the prediction effect of the detection model. Specifically, the ratio of the number of training samples to prediction samples may be 9:1. Test samples can be derived from CASIA 1.0 and Columbia datasets.

Step 104: Based on the trained detection model, determine whether the detection image has been tampered with.

The embodiment of the present invention performs boundary detection and area detection on the image based on the detection model. The area detection is based on the feature difference between the tampered area and the background area of the entire image, and the tampered area is identified, focusing on the overall characteristics of the image; the boundary detection is based on the two sides of the tampered boundary. feature differences, identify tampered boundaries, and focus on local features. Boundary detection can assist region detection, determine the tampered region more accurately, and improve the accuracy of image detection.

In an embodiment of the present invention, the detection model includes: a feature extraction layer, a region detection layer, and a boundary detection layer;

Input the training image into the detection model to obtain the region detection results and boundary detection results, including:

Input the training image into the feature extraction layer to extract high-order feature maps and low-order feature maps from the training images;

Input the high-order feature map and low-order feature map into the region detection layer to obtain the region detection result;

Input the high-order feature map and low-order feature map into the boundary detection layer to obtain the boundary detection result.

The embodiment of the present invention determines the region detection result and the boundary detection result based on the detection model. Based on different functions, the detection model can be divided into feature extraction layer, region detection layer and boundary detection layer. Among them, the feature extraction layer is used to extract high-order features and low-order features from the training images, the extracted high-order features form a high-order feature map, and the extracted low-order features form a low-order feature map. Low-order features have higher resolution, including location information, detail information, etc.; high-order features have more semantic information, but lower resolution. The region detection layer is used to detect tampered regions, and the boundary detection layer is used to detect tampered boundaries.

The embodiment of the present invention uses high-order features and low-order features to perform boundary detection and region detection, and considers multi-dimensional features in training images to improve the accuracy of region detection and boundary detection, thereby tampering with the accuracy of identification results.

In one embodiment of the present invention, the training image is input into the feature extraction layer to extract high-order feature maps and low-order feature maps from the training images, including:

Input the training image into the backbone network to obtain the low-level feature map and the first feature map;

Extracting multi-scale features from the first feature map based on the multi-scale network to obtain multiple second feature maps;

After splicing multiple second feature maps and inputting them into the first convolutional layer, a high-order feature map is obtained;

The backbone network includes: a first multi-channel convolutional layer and a depthwise separable convolutional layer; the first convolutional layer is a 1×1 convolutional layer.

In the embodiment of the present invention, the feature extraction layer includes a backbone network, a multi-scale network and a first convolution layer. The backbone network is used to extract features from training images, which can be implemented by multi-channel convolutional layers and depthwise separable convolutions. The combination of the first multi-channel convolutional layer and the depthwise separable convolutional layer can improve the efficiency and quality of feature extraction. In this embodiment of the present invention, the backbone network may include multiple first multi-channel convolutional layers and multiple depthwise separable convolutional layers, and may also replace the depthwise separable convolutional layers with the first multi-channel convolutional layer or Other types of convolutional layers.

The embodiments of the present invention improve the accuracy of region detection by extracting multi-scale features, thereby improving the accuracy and reliability of image detection results.

The first convolutional layer is used to fuse the second feature maps to obtain high-order feature maps.

In one embodiment of the present invention, the multi-scale network includes: an atrous convolutional layer, a second convolutional layer and a pooling layer;

Among them, the second convolutional layer is a 1×1 convolutional layer.

In the embodiment of the present invention, the receptive field is enlarged by the hole convolution layer, so that the multi-scale network can output more abundant information, thereby improving the model training effect. In this embodiment of the present invention, multiple atrous convolutional layers, such as three or four, may be used.

In an embodiment of the present invention, the region detection layer includes: a first feature fusion layer, a region anomaly analysis layer, and a first result output layer;

Input the high-order feature map and low-order feature map into the region detection layer to obtain the region detection results, including:

Input the high-order feature map and the low-order feature map into the first feature fusion layer to obtain the third feature map;

Determine a fourth feature map according to the third feature map and the regional anomaly analysis layer; wherein, the fourth feature map is used to represent the pixel value difference between the tampered area and the background area in the third feature map;

The fourth feature map is input into the first result output layer to obtain the region detection result.

The embodiment of the present invention determines the range of the tampered area based on the pixel value difference between the tampered area and the background area. In the process of calculating the pixel value difference, the pixel value of each pixel coordinate in the third feature map is considered, and the tampered area can be identified from a global perspective.

In an embodiment of the present invention, the high-order feature map and the low-order feature map are input into the first feature fusion layer to obtain a third feature map, including:

Input the low-level feature map into the third convolutional layer to obtain the fifth feature map;

Upsampling the high-order feature map to obtain the sixth feature map;

The fifth feature map and the sixth feature map are spliced and input into the second multi-channel convolutional layer to obtain the third feature map;

Among them, the third convolutional layer is a 1×1 convolutional layer.

The embodiment of the present invention performs feature fusion and compression on the low-level feature map through a 1×1 convolution layer, so as to remove redundant features and improve the model training effect. In addition, higher-order feature maps can be upsampled by bilinear interpolation or transposed convolution to enlarge the size of higher-order feature maps. The fifth feature map and the sixth feature map can be spliced according to the Z axis, and then feature fusion is performed through a 1×1 convolutional layer.

In an embodiment of the present invention, the fourth feature map is determined according to the third feature map and the regional anomaly analysis layer, including:

Calculate the average pixel value of the third feature map according to the pixel value of each pixel coordinate in the third feature map;

Determine the difference between the pixel value of each pixel coordinate and the average pixel value;

Calculate the pixel value standard deviation of the third feature map according to the difference between the pixel value of each pixel coordinate and the average pixel value;

Calculate the normalized pixel value of each pixel coordinate according to the standard deviation of the pixel value and the difference between the pixel value of each pixel coordinate and the average pixel value;

A fourth feature map is determined according to the normalized pixel values of the respective pixel coordinates.

In this embodiment of the present invention, the normalized pixel value is used to represent the degree of difference between the pixel value of the pixel coordinate and the average pixel value of the third feature map. The embodiment of the present invention identifies whether the pixel coordinates are in the tampered area through the difference of pixel values, which can improve the accuracy of area identification.

In an embodiment of the present invention, the fourth feature map is input into the first result output layer to obtain a region detection result, including:

Input the fourth feature map into the fourth convolution layer to obtain the seventh feature map;

Upsampling the seventh feature map to obtain the eighth feature map;

Input the eighth feature map into the activation function to obtain the region detection result;

Among them, the fourth convolutional layer is a 1×1 convolutional layer.

In the embodiment of the present invention, the image is enlarged by up-sampling, and the pixel value is mapped between 0 and 1 through the activation function, and the mapping result of each pixel coordinate of the eighth feature map constitutes the region detection result. In the embodiment of the present invention, dimension reduction and amplification are performed after the normalized pixel value is calculated, which can ensure the accuracy of the pixel value used in the calculation process and improve the accuracy and reliability of the area detection result. The activation function used can be a sigmoid function, a softmax function, or the like.

In an embodiment of the present invention, the boundary detection layer includes: a second feature fusion layer, a boundary anomaly analysis layer, and a second result output layer;

Input the high-order feature map and low-order feature map into the boundary detection layer to obtain boundary detection results, including:

Input the high-order feature map and the low-order feature map into the second feature fusion layer to obtain the ninth feature map;

According to the ninth feature map and the boundary anomaly analysis layer, determine the tenth feature map; wherein, the tenth feature map is used to represent the pixel value difference between the tampered area and the background area in the detection window;

The tenth feature map is input into the second result output layer to obtain the boundary detection result.

The embodiment of the present invention determines the range of the tampered area based on the pixel value difference between the tampered area and the background area in the detection window. Different from area detection, the embodiment of the present invention considers the pixel value of each pixel coordinate in the detection window, and can assist the area detection process to determine the tampered area from a local perspective.

In an embodiment of the present invention, the high-order feature map and the low-order feature map are input into the second feature fusion layer to obtain a ninth feature map, including:

Input the low-level feature map into the fifth convolutional layer to obtain the eleventh feature map;

Upsampling the high-order feature map to obtain the twelfth feature map;

The eleventh feature map and the twelfth feature map are spliced and input into the third multi-channel convolutional layer to obtain the ninth feature map;

Among them, the fifth convolutional layer is a 1×1 convolutional layer.

Similar to the region detection part, the embodiment of the present invention performs feature fusion and compression on the low-level feature map through a 1×1 convolution layer, so as to remove redundant features and improve the model training effect. In addition, higher-order feature maps can be upsampled by bilinear interpolation or transposed convolution to enlarge the size of higher-order feature maps. The eleventh feature map and the twelfth feature map can be spliced according to the Z axis, and then feature fusion is performed through a 1×1 convolution layer.

In an embodiment of the present invention, the tenth feature map is determined according to the ninth feature map and the boundary anomaly analysis layer, including:

Calculate the average pixel value of the detection window according to the pixel value of each pixel coordinate in the ninth feature map in the detection window;

Determine the difference between the pixel value of each pixel coordinate and the average pixel value of the detection window where the pixel coordinate is located;

Calculate the pixel value standard deviation of the ninth feature map;

According to the standard deviation of the pixel value, the difference between the pixel value of each pixel coordinate and the average pixel value of the detection window where the pixel coordinate is located, calculate the normalized pixel value of the pixel coordinate in the detection window;

The tenth feature map is determined according to the normalized pixel values of the pixel coordinates within the detection window.

The embodiment of the present invention focuses on the pixel value difference in the local area of the detection window. There is a difference in pixel values on both sides of the tampering boundary, which can be determined by computing the normalized pixel values of the pixel coordinates within the detection window. In practical application scenarios, the size of the detection window can be adjusted as required.

In an embodiment of the present invention, the tenth feature map is input into the second result output layer to obtain a boundary detection result, including:

Input the tenth feature map into the sixth convolutional layer to obtain the thirteenth feature map;

Upsampling the thirteenth feature map to obtain the fourteenth feature map;

Input the fourteenth feature map into the activation function to obtain the region detection result;

Among them, the sixth convolutional layer is a 1×1 convolutional layer.

The second result output layer is similar to the first result output layer. In the embodiment of the present invention, the image is enlarged by up-sampling, and the pixel value is mapped between 0 and 1 through the activation function. The mapping result of each pixel coordinate of the fourteenth feature map is composed of Area detection results. The activation function used can be a sigmoid function, a softmax function, or the like.

In one embodiment of the present invention, acquiring training samples includes:

Get training images and region labels;

Perform an expansion operation on the region label to obtain an expanded image;

Perform the erosion operation on the region label to get the erosion image;

Based on the dilated and eroded images, the boundary labels are determined.

In the embodiment of the present invention, since there is no boundary label in CASIA 2.0, the embodiment of the present invention generates a boundary label based on the area label. The difference between the dilated image and the eroded image is the boundary label. Dilation and erosion operations can be implemented with a 7x7 window. Through the embodiment of the present invention, the boundary label can be obtained more conveniently, and the model training efficiency can be improved.

In an embodiment of the present invention, the method further includes: acquiring pre-training samples; pre-training the detection model based on the pre-training samples;

Input the training image into the detection model to obtain the region detection results and boundary detection results, including:

Input the training image into the pre-trained detection model to obtain the region detection results and boundary detection results.

In this embodiment of the present invention, the pre-training samples may be constructed from the original images in the COCO dataset. For example, select an image in the COCO dataset as the original image, then crop an object from another image, and paste it into the original image through operations such as rotation, enlargement, etc. The embodiment of the present invention improves the training effect of the detection model through pre-training, thereby improving the accuracy of tampered image detection.

As shown in FIG. 2, an embodiment of the present invention provides an image detection method, including:

Step 201: Obtain pre-training samples.

Select the original image from the COCO dataset, crop the object image from another image, and paste the object image into the original image after rotation and method to obtain pre-training samples.

Step 202: Pre-train the detection model based on the pre-training samples.

The architecture of the detection model is shown in FIG. 3 , and the following embodiments will describe the architecture in detail.

Step 203: Obtain training images and region labels.

Get training images and region labels from CASIA 2.0.

Step 204: Perform an expansion operation on the region label to obtain an expanded image.

Step 205: Perform an erosion operation on the region label to obtain an erosion image.

The window size used for dilation and erosion operations is 7 × 7.

As shown in Figure 4, from left to right are region labels, dilated images, eroded images, and boundary labels.

Step 206: Determine boundary labels according to the dilated image and the eroded image.

Training images, region labels, and boundary labels constitute training samples.

Step 207: Input the training image into the feature extraction layer to extract high-order feature maps and low-order feature maps from the training images.

Specifically, the training images are input into the backbone network. The structure of the backbone network is shown in Figure 5. As can be seen from the figure, the backbone network includes an entry layer, an intermediate layer and an exit layer. The entry layer consists of five multi-channel convolutional layers and nine depthwise separable convolutional layers. Taking "Conv 32, 3x3, stride2" as an example, Conv 32 means that the output channel of the multi-channel convolutional layer is 32, the convolution kernel is 3x3, and the stride is 2. The middle layer consists of 16 identical depthwise separable convolutional layers. The exit layer consists of one multi-channel convolutional layer and six depthwise separable convolutional layers. The low-cost feature map is output by the third depthwise separable convolutional layer in the middle layer, and the first feature map is output by the export layer.

Referring to Figure 3, the first feature map is sequentially input into the convolution kernel of 3x3, three atrous convolutional layers with expansion rates of 6, 12, and 18, a 1x1 convolutional layer and a pooling layer to obtain multiple second features. picture. The second feature map is spliced along the Z axis and input into a 1x1 convolutional layer to obtain a high-order feature map that fuses features of different scales. In the embodiment of the present invention, the size of the low-level feature map is 1/4 of the training image, and the size of the high-level feature map is 1/16 of the training image. The multi-scale network can also be the structure shown in Figure 6, which includes a 1x1 convolutional layer and dilated convolutions with dilation rates of 1, 2, and 5. In Figure 6, each horizontal convolutional layer shares convolution kernel parameters, so that the same target has the same feature expression ability at different scales.

Step 208: Input the high-order feature map and the low-order feature map into the region detection layer to obtain a region detection result.

Specifically, the low-order feature map is input into the 1x1 convolutional layer to obtain the fifth feature map. Perform bilinear interpolation on the higher-order feature map to make it quadruple to obtain the sixth feature map. Splicing the fifth feature map and the sixth feature map according to the Z axis, and then inputting the 3x3 convolution layer to fuse the features to obtain the third feature map.

According to the pixel value of each pixel coordinate in the third feature map, the average pixel value of the third feature map is calculated, as shown in formula (1).

Among them, F[i,j] is used to represent the pixel value of the pixel coordinate (i, j) in the third feature map, H is used to represent the height of the third feature map, W is used to represent the width of the third feature map, μ _f is used to characterize the average pixel value of the third feature map.

Determine the difference between the pixel value of each pixel coordinate and the average pixel value, as shown in equation (2).

D _f [i,j]=F[i,j]-μ _f (2)

Among them, D _f [i, j] is used to represent the difference between the pixel value of the pixel coordinate (i, j) and the average pixel value.

According to the difference between the pixel value of each pixel coordinate and the average pixel value, the pixel value standard deviation of the third feature map is calculated.

According to the standard deviation of the pixel value and the difference between the pixel value of each pixel coordinate and the average pixel value, the normalized pixel value of each pixel coordinate is calculated, as shown in formula (3).

Z _f [i,j]=D _f [i,j]/max(σ _f ,ε+ω _σ1 ) (3)

Among them, σ _f is used to represent the pixel value standard deviation of the third feature map, ε is 10 ^-5 , and ω _σ1 is a first vector that can be continuously adjusted through the training process.

A fourth feature map is determined according to the normalized pixel values of the respective pixel coordinates.

Input the fourth feature map into a 1×1 convolutional layer to get the seventh feature map. The seventh feature map is enlarged four times by bilinear interpolation to obtain the eighth feature map. Input the eighth feature map into the sigmoid function to obtain the region detection result.

Step 209: Input the high-order feature map and the low-order feature map into the boundary detection layer to obtain a boundary detection result.

Specifically, the low-order feature map is input into the 1x1 convolutional layer to obtain the eleventh feature map. Perform bilinear interpolation on the higher-order feature map to make it quadruple to obtain the twelfth feature map. The eleventh feature map and the twelfth feature map are spliced according to the Z axis, and then the 3x3 convolution layer is input to fuse the features to obtain the ninth feature map.

According to the pixel value of each pixel coordinate in the ninth feature map in the detection window, the average pixel value of the detection window is calculated, as shown in formula (4).

用于表征高为7、宽为7的检测窗口的平均像素值。in,

The average pixel value used to characterize a detection window with a height of 7 and a width of 7.

Determine the difference between the pixel value of each pixel coordinate and the average pixel value of the detection window where the pixel coordinate is located, as shown in formula (5).

用于表征像素坐标(i，j)的像素值与像素坐标所处检测窗口的平均像素值的差。in,

The difference between the pixel value used to characterize the pixel coordinate (i, j) and the average pixel value of the detection window where the pixel coordinate is located.

Calculate the standard deviation of pixel values of the ninth feature map.

According to the standard deviation of the pixel value, the difference between the pixel value of each pixel coordinate and the average pixel value of the detection window where the pixel coordinate is located, the normalized pixel value of the pixel coordinate in the detection window is calculated, as shown in formula (6).

In the embodiment of the present invention, the ninth feature map is the same as the third feature map, and the standard deviations of the pixel values of the two are the same. _ωσ2 is a second vector that can be continuously adjusted through the training process.

The tenth feature map is determined according to the normalized pixel values of the pixel coordinates within the detection window.

Input the tenth feature map into a 1×1 convolutional layer to get the thirteenth feature map. The thirteenth feature map is enlarged by a factor of four by bilinear interpolation to obtain the fourteenth feature map. Input the fourteenth feature map into the sigmoid function to get the boundary detection result.

Step 210: Train the detection model according to the region label, the boundary label, the region detection result and the boundary detection result.

According to the difference between the area label and the area detection result, the difference between the predicted tampering area and the actual tampering area can be determined; according to the difference between the boundary label and the boundary detection result, the difference between the predicted tampering boundary and the actual tampering boundary can be determined. The embodiment of the present invention adopts a cross-entropy loss function, which includes two parts: region detection and boundary detection, as shown in equations (7)-(9).

用于表征训练样本k的区域检测结果，

用于表征训练样本k的边界检测结果，

用于表征训练样本k中像素坐标(i，j)对应的区域标签的值，

用于表征训练样本k中像素坐标(i，j)的区域检测结果，

用于表征训练样本k中像素坐标(i，j)对应的边界标签的值，

用于表征训练样本k中像素坐标(i，j)的边界检测结果。Among them, m is used to represent the number of training samples,

is used to characterize the region detection results of training sample k,

is used to characterize the boundary detection results of the training sample k,

is used to characterize the value of the region label corresponding to the pixel coordinate (i, j) in the training sample k,

is used to characterize the region detection results of the pixel coordinates (i, j) in the training sample k,

is used to characterize the value of the boundary label corresponding to the pixel coordinate (i, j) in the training sample k,

Used to characterize the boundary detection results of pixel coordinates (i, j) in training sample k.

The loss value can be calculated by formulas (7)-(9), and the parameters of the detection model can be adjusted according to the loss value.

Step 211: Based on the trained detection model, determine whether the detection image has been tampered with.

The detection model can map the input pixel value between 0 and 1. If the value output by the Sigmoid function is greater than the set value (0.5 in the embodiment of the present invention), it is determined that the pixel coordinates are located in the tampering area, otherwise, it is located in the background area. .

In the embodiment of the present invention, Columbia and CASIA 1.0 are used as test sample sets, and the performance of the detection model obtained by training is evaluated by the F1 score. The test results are shown in Table 1. It can be seen from Table 1 that, compared with other models, the detection model trained in the embodiment of the present invention has the highest F1 score, indicating that its performance is better than other models. Among them, RGB-N is a tampered image detection method based on dual-stream Faster R-CNN, NOI1 is a method for detecting tampered images based on noise inconsistency, which uses high-pass wavelet coefficients to simulate local noise, and CFA is a CFA mode Estimation method, which uses nearby pixels to approximate the camera filter array pattern and then produces a tampering probability for each pixel. DCT is a JPEG image forgery detection method based on the histogram difference of DCT coefficients.

Table 1 F1 scores of different models

Columbia CASIA 1.0 Detection model 0.747 0.435 RGB-N 0.697 0.408 NOI1 0.574 0.263 DCT 0.520 0.301 CFA 0.503 0.212

As shown in FIG. 7 , an embodiment of the present invention provides an image detection apparatus, including:

The obtaining module 701 is configured to obtain training samples; wherein, the training samples include: training images, region labels and boundary labels;

The training module 702 is configured to input the training image into the detection model to obtain the region detection result and the boundary detection result; train the detection model according to the region label, the boundary label, the region detection result and the boundary detection result;

The detection module 703 is configured to determine whether the detected image has been tampered with based on the trained detection model.

In an embodiment of the present invention, the detection model includes: a feature extraction layer, a region detection layer, and a boundary detection layer;

The training module 702 is configured to input the training image into the feature extraction layer, so as to extract the high-order feature map and the low-order feature map from the training image; input the high-order feature map and the low-order feature map into the region detection layer to obtain the region detection result ; Input the high-order feature map and low-order feature map into the boundary detection layer to obtain the boundary detection result.

In an embodiment of the present invention, the training module 702 is configured to input the training image into the backbone network to obtain a low-order feature map and a first feature map; extract multi-scale features from the first feature map based on a multi-scale network, and obtain a a second feature map; multiple second feature maps are spliced and input into the first convolutional layer to obtain a high-order feature map; wherein, the backbone network includes: a first multi-channel convolutional layer and a depthwise separable convolutional layer; A convolutional layer is a 1×1 convolutional layer.

In an embodiment of the present invention, the multi-scale network includes: an atrous convolutional layer, a second convolutional layer, and a pooling layer; wherein the second convolutional layer is a 1×1 convolutional layer.

In an embodiment of the present invention, the region detection layer includes: a first feature fusion layer, a region anomaly analysis layer, and a first result output layer; the training module 702 is configured to input the high-order feature map and the low-order feature map into the first result output layer. a feature fusion layer to obtain a third feature map; according to the third feature map and the regional anomaly analysis layer, a fourth feature map is determined; wherein the fourth feature map is used to represent the pixel values of the tampered area and the background area in the third feature map difference; input the fourth feature map into the first result output layer to obtain the region detection result.

In an embodiment of the present invention, the training module 702 is configured to input the low-order feature map into the third convolutional layer to obtain the fifth feature map; upsample the high-order feature map to obtain the sixth feature map; The fifth feature map and the sixth feature map are spliced and input into the second multi-channel convolutional layer to obtain the third feature map; wherein, the third convolutional layer is a 1×1 convolutional layer.

In an embodiment of the present invention, the training module 702 is configured to calculate the average pixel value of the third feature map according to the pixel value of each pixel coordinate in the third feature map; determine the difference between the pixel value of each pixel coordinate and the average pixel value difference; according to the difference between the pixel value of each pixel coordinate and the average pixel value, calculate the pixel value standard deviation of the third feature map; according to the pixel value standard deviation, the difference between the pixel value of each pixel coordinate and the average pixel value, calculate each pixel coordinate The normalized pixel value of , and the fourth feature map is determined according to the normalized pixel value of each pixel coordinate.

In an embodiment of the present invention, the training module 702 is configured to input the fourth feature map into the fourth convolution layer to obtain the seventh feature map; upsample the seventh feature map to obtain the eighth feature map; The eight feature maps are input to the activation function to obtain the region detection results; among them, the fourth convolutional layer is a 1×1 convolutional layer.

In an embodiment of the present invention, the boundary detection layer includes: a second feature fusion layer, a boundary abnormality analysis layer, and a second result output layer; the training module 702 is configured to input the high-order feature map and the low-order feature map into the first The second feature fusion layer is to obtain the ninth feature map; the tenth feature map is determined according to the ninth feature map and the boundary anomaly analysis layer; wherein, the tenth feature map is used to represent the pixel value difference between the tampered area and the background area in the detection window; The tenth feature map is input into the second result output layer to obtain the boundary detection result.

In an embodiment of the present invention, the training module 702 is configured to input the low-order feature map into the fifth convolutional layer to obtain an eleventh feature map; perform upsampling on the high-order feature map to obtain the twelfth feature map; The eleventh feature map and the twelfth feature map are spliced and input into the third multi-channel convolutional layer to obtain the ninth feature map; among them, the fifth convolutional layer is a 1×1 convolutional layer.

In an embodiment of the present invention, the training module 702 is configured to calculate the average pixel value of the detection window according to the pixel value of each pixel coordinate in the ninth feature map in the detection window; Calculate the standard deviation of the pixel values of the ninth feature map; calculate the detection window according to the standard deviation of the pixel values, the difference between the pixel value of each pixel coordinate and the average pixel value of the detection window where the pixel coordinates are located The normalized pixel value of the inner pixel coordinates; the tenth feature map is determined according to the normalized pixel value of the pixel coordinates in the detection window.

In an embodiment of the present invention, the training module 702 is configured to input the tenth feature map into the sixth convolutional layer to obtain the thirteenth feature map; and upsample the thirteenth feature map to obtain the fourteenth feature map ; Input the fourteenth feature map into the activation function to obtain the region detection result; among them, the sixth convolutional layer is a 1×1 convolutional layer.

In an embodiment of the present invention, the acquisition module 701 is configured to acquire training images and region labels; perform a dilation operation on the region labels to obtain a dilated image; perform an erosion operation on the region labels to obtain an eroded image; , to determine the boundary labels.

In an embodiment of the present invention, the acquisition module 701 is configured to acquire pre-training samples; pre-train the detection model based on the pre-training samples; input the training image into the pre-trained detection model to obtain the region detection result and the boundary detection result .

An embodiment of the present invention provides an electronic device, including:

one or more processors;

storage means for storing one or more programs,

When one or more programs are executed by one or more processors, the one or more processors implement the method as described in any of the above embodiments.

An embodiment of the present invention provides a computer-readable medium on which a computer program is stored, and when the program is executed by a processor, the method described in any of the foregoing embodiments is implemented.

FIG. 8 shows an exemplary system architecture 800 to which an image detection method or an image detection apparatus according to an embodiment of the present invention may be applied.

As shown in FIG. 8 , the system architecture 800 may include terminal devices 801 , 802 , and 803 , a network 804 and a server 805 . The network 804 is a medium used to provide a communication link between the terminal devices 801 , 802 , 803 and the server 805 . Network 804 may include various connection types, such as wired, wireless communication links, or fiber optic cables, among others.

The user can use the terminal devices 801, 802, 803 to interact with the server 805 through the network 804 to receive or send messages and the like. Various communication client applications may be installed on the terminal devices 801 , 802 and 803 , such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only examples).

The terminal devices 801, 802, 803 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

The server 805 may be a server that provides various services, such as a background management server that provides support for shopping websites browsed by the terminal devices 801 , 802 and 803 (just an example). The background management server can analyze and process the received product information query request and other data, and feed back the processing results (such as target push information, product information—just an example) to the terminal device.

It should be noted that the image detection method provided in the embodiment of the present invention is generally executed by the server 805 , and accordingly, the image detection apparatus is generally set in the server 805 .

It should be understood that the numbers of terminal devices, networks and servers in FIG. 8 are merely illustrative. There can be any number of terminal devices, networks and servers according to implementation needs.

Referring next to FIG. 9 , it shows a schematic structural diagram of a computer system 900 suitable for implementing a terminal device according to an embodiment of the present invention. The terminal device shown in FIG. 9 is only an example, and should not impose any limitations on the functions and scope of use of the embodiments of the present invention.

As shown in FIG. 9, a computer system 900 includes a central processing unit (CPU) 901, which can be loaded into a random access memory (RAM) 903 according to a program stored in a read only memory (ROM) 902 or a program from a storage section 908 Instead, various appropriate actions and processes are performed. In the RAM 903, various programs and data necessary for the operation of the system 900 are also stored. The CPU 901 , the ROM 902 , and the RAM 903 are connected to each other through a bus 904 . An input/output (I/O) interface 905 is also connected to bus 904 .

The following components are connected to the I/O interface 905: an input section 906 including a keyboard, a mouse, etc.; an output section 907 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 908 including a hard disk, etc. ; and a communication section 909 including a network interface card such as a LAN card, a modem, and the like. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the I/O interface 905 as needed. A removable medium 911, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is mounted on the drive 910 as needed so that a computer program read therefrom is installed into the storage section 908 as needed.

In particular, the processes described above with reference to the flowcharts may be implemented as computer software programs in accordance with the disclosed embodiments of the present invention. For example, embodiments disclosed herein include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the method illustrated in the flowchart. In such an embodiment, the computer program may be downloaded and installed from the network via the communication portion 909, and/or installed from the removable medium 911. When the computer program is executed by the central processing unit (CPU) 901, the above-described functions defined in the system of the present invention are executed.

It should be noted that the computer-readable medium shown in the present invention may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or a combination of any of the above. More specific examples of computer readable storage media may include, but are not limited to, electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read only memory (ROM), erasable Programmable read only memory (EPROM or flash memory), fiber optics, portable compact disk read only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing. In the present invention, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In the present invention, however, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code therein. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device . Program code embodied on a computer readable medium may be transmitted using any suitable medium including, but not limited to, wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code that contains one or more logical functions for implementing the specified functions executable instructions. It should also be noted that, in some alternative implementations, the functions noted in the blocks may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It is also noted that each block of the block diagrams or flowchart illustrations, and combinations of blocks in the block diagrams or flowchart illustrations, can be implemented in special purpose hardware-based systems that perform the specified functions or operations, or can be implemented using A combination of dedicated hardware and computer instructions is implemented.

The modules involved in the embodiments of the present invention may be implemented in a software manner, and may also be implemented in a hardware manner. The described modules can also be provided in the processor, for example, it can be described as: a processor includes a sending module, an obtaining module, a determining module and a first processing module. Among them, the names of these modules do not constitute a limitation of the module itself in some cases, for example, the sending module can also be described as "a module that sends a request for image acquisition to the connected server".

As another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiments; or may exist alone without being assembled into the device. The above-mentioned computer-readable medium carries one or more programs, and when the above-mentioned one or more programs are executed by a device, the device includes:

Obtain training samples; wherein, the training samples include: training images, region labels and boundary labels;

Inputting the training image into a detection model to obtain a region detection result and a boundary detection result;

train the detection model according to the region label, the boundary label, the region detection result and the boundary detection result;

Based on the trained detection model, it is determined whether the detection image has been tampered with.

According to the technical solutions of the embodiments of the present invention, boundary detection and area detection are performed on the image based on the detection model, and the area detection is based on the feature difference between the tampered area and the background area of the entire image, and the tampered area is identified, which focuses on the overall characteristics of the image; boundary detection Based on the feature difference on both sides of the tampered boundary, the tampered boundary is identified. Boundary detection can assist region detection, determine the tampered region more accurately, and improve the accuracy of image detection.

The above-mentioned specific embodiments do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims (17)

1. An image detection method, comprising:

obtaining a training sample; wherein the training samples comprise: training images, area labels and boundary labels;

inputting the training image into a detection model to obtain an area detection result and a boundary detection result;

training the detection model according to the area label, the boundary label, the area detection result and the boundary detection result;

and determining whether the detection image is tampered or not based on the trained detection model.

2. The method of claim 1,

the detection model comprises: a feature extraction layer, a region detection layer and a boundary detection layer;

inputting the training image into the detection model to obtain a region detection result and a boundary detection result, wherein the method comprises the following steps:

inputting the training image into the feature extraction layer to extract a high-order feature map and a low-order feature map from the training image;

inputting the high-order characteristic diagram and the low-order characteristic diagram into the area detection layer to obtain the area detection result;

and inputting the high-order characteristic diagram and the low-order characteristic diagram into the boundary detection layer to obtain the boundary detection result.

3. The method of claim 2,

the inputting the training image into the feature extraction layer to extract a high-order feature map and a low-order feature map from the training image includes:

inputting the training image into a backbone network to obtain the low-order feature map and a first feature map;

extracting multi-scale features from the first feature map based on a multi-scale network to obtain a plurality of second feature maps;

splicing the second feature maps and inputting the second feature maps into a first convolution layer to obtain the high-order feature map;

wherein the backbone network comprises: a first multi-channel convolutional layer and a depth-separable convolutional layer; the first convolutional layer is a 1 × 1 convolutional layer.

4. The method of claim 3,

the multi-scale network comprises: a void convolutional layer, a second convolutional layer and a pooling layer;

wherein the second convolutional layer is a 1 × 1 convolutional layer.

5. The method of claim 2,

the area detection layer includes: the system comprises a first feature fusion layer, a regional anomaly analysis layer and a first result output layer;

the inputting the high-order feature map and the low-order feature map into the area detection layer to obtain the area detection result includes:

inputting the high-order feature map and the low-order feature map into the first feature fusion layer to obtain a third feature map;

determining a fourth feature map according to the third feature map and the regional anomaly analysis layer; the fourth feature map is used for representing the difference of pixel values of the tampered area and the background area in the third feature map;

and inputting the fourth feature map into the first result output layer to obtain the region detection result.

6. The method of claim 5,

inputting the high-order feature map and the low-order feature map into the first feature fusion layer to obtain a third feature map, including:

inputting the low-order feature map into a third convolutional layer to obtain a fifth feature map;

the high-order characteristic diagram is subjected to up-sampling to obtain a sixth characteristic diagram;

splicing the fifth feature map and the sixth feature map, and inputting the spliced fifth feature map and sixth feature map into a second multi-channel convolutional layer to obtain a third feature map;

wherein the third convolutional layer is a 1 × 1 convolutional layer.

7. The method of claim 5,

determining a fourth feature map according to the third feature map and the regional anomaly analysis layer, wherein the determining comprises:

calculating an average pixel value of the third feature map according to the pixel value of each pixel coordinate in the third feature map;

determining a difference between a pixel value of each of the pixel coordinates and the average pixel value;

calculating a pixel value standard deviation of the third feature map according to the difference between the pixel value of each pixel coordinate and the average pixel value;

calculating a normalized pixel value of each pixel coordinate according to the standard deviation of the pixel values and the difference between the pixel value of each pixel coordinate and the average pixel value;

and determining the fourth feature map according to the normalized pixel value of each pixel coordinate.

8. The method according to any one of claims 5 to 7,

the inputting the fourth feature map into the first result output layer to obtain the region detection result includes:

inputting the fourth feature map into a fourth convolutional layer to obtain a seventh feature map;

up-sampling the seventh characteristic diagram to obtain an eighth characteristic diagram;

inputting the eighth feature map into an activation function to obtain the region detection result;

wherein the fourth convolutional layer is a 1 × 1 convolutional layer.

9. The method of claim 2,

the boundary detection layer includes: the second characteristic fusion layer, the boundary anomaly analysis layer and the second result output layer;

the inputting the high-order feature map and the low-order feature map into the boundary detection layer to obtain the boundary detection result includes:

inputting the high-order feature map and the low-order feature map into the second feature fusion layer to obtain a ninth feature map;

determining a tenth feature map according to the ninth feature map and the boundary anomaly analysis layer; wherein the tenth feature map is used for characterizing the difference of pixel values of the tampered region and the background region in the detection window;

and inputting the tenth feature map into the second result output layer to obtain the boundary detection result.

10. The method of claim 9,

inputting the high-order feature map and the low-order feature map into the second feature fusion layer to obtain a ninth feature map, including:

inputting the low-order feature map into a fifth convolutional layer to obtain an eleventh feature map;

performing upsampling on the high-order characteristic diagram to obtain a twelfth characteristic diagram;

splicing the eleventh characteristic diagram and the twelfth characteristic diagram, and inputting the spliced eleventh characteristic diagram and twelfth characteristic diagram into a third multi-channel convolutional layer to obtain a ninth characteristic diagram;

wherein the fifth convolutional layer is a 1 × 1 convolutional layer.

11. The method of claim 9,

determining a tenth feature map according to the ninth feature map and the boundary anomaly analysis layer, including:

calculating the average pixel value of the detection window according to the pixel value of each pixel coordinate in the ninth characteristic diagram in the detection window;

determining the difference between the pixel value of each pixel coordinate and the average pixel value of the detection window where the pixel coordinate is located;

calculating a pixel value standard deviation of the ninth feature map;

calculating a normalized pixel value of a pixel coordinate in a detection window according to the standard deviation of the pixel value, the difference between the pixel value of each pixel coordinate and the average pixel value of the detection window where the pixel coordinate is located;

and determining the tenth feature map according to the normalized pixel value of the pixel coordinate in the detection window.

12. The method according to any one of claims 9 to 11,

inputting the tenth feature map into the second result output layer to obtain the boundary detection result, where the method includes:

inputting the tenth characteristic diagram into a sixth convolutional layer to obtain a thirteenth characteristic diagram;

performing upsampling on the thirteenth characteristic diagram to obtain a fourteenth characteristic diagram;

inputting the fourteenth feature map into an activation function to obtain the region detection result;

wherein the sixth convolutional layer is a 1 × 1 convolutional layer.

13. The method of claim 1,

the obtaining of the training sample includes:

acquiring the training image and the area label;

performing expansion operation on the area label to obtain an expansion image;

performing corrosion operation on the area label to obtain a corrosion image;

and determining the boundary label according to the expansion image and the erosion image.

14. The method of claim 1, further comprising:

obtaining a pre-training sample;

pre-training the detection model based on the pre-training samples;

inputting the training image into a detection model to obtain a region detection result and a boundary detection result, wherein the method comprises the following steps:

and inputting the training image into the pre-trained detection model to obtain the region detection result and the boundary detection result.

15. An image detection apparatus, characterized by comprising:

an acquisition module configured to acquire a training sample; wherein the training samples comprise: training images, area labels and boundary labels;

the training module is configured to input the training image into a detection model to obtain an area detection result and a boundary detection result; training the detection model according to the area label, the boundary label, the area detection result and the boundary detection result;

and the detection module is configured to determine whether the detection image is tampered based on the trained detection model.

16. An electronic device, comprising:

one or more processors;

a storage device for storing one or more programs,

when executed by the one or more processors, cause the one or more processors to implement the method of any one of claims 1-14.

17. A computer-readable medium, on which a computer program is stored, which, when being executed by a processor, carries out the method according to any one of claims 1-14.

Images