False video detection method and false video detection device using the same
A false video detection method is provided. The method includes: retrieving N video segments in a target video, wherein each video segment includes M image frames; performing a feature retrieving procedure on the N×M image frames to obtain feature maps corresponding to the N×M image frames; inputting the feature maps into a time-attention model to obtain concentrated feature; multiplying the concentrated feature with the feature maps to obtain first enhanced feature maps, adding the first enhanced feature maps with the feature maps to obtain second enhanced feature maps, and inputting the second enhanced feature maps into a fully connected layer to obtain a determination result outputted by the fully connected layer, wherein the determination result indicates that the target video is true or false.
1 . A false video detection method, performed by a computing device comprising a processor, the method comprising:
extracting, by the processor, N video segments from a target video, wherein each video segment contains M image frames;
performing, by the processor, a feature extraction procedure on the N×M image frames to obtain a feature map corresponding to the N×M image frames;
inputting, by the processor, the feature map into a time attention model to obtain a concentrated feature;
performing, by the processor, a multiplication operation to the concentrated feature and the feature map to obtain a first enhanced feature map;
performing, by the processor, a summing operation on the first enhanced feature map and the feature map to obtain a second enhanced feature map; and
inputting, by the processor, the second enhanced feature map into a fully connected layer to obtain a judgment result outputted by the fully connected layer, wherein the judgment result indicates that the target video is true or false.
2 . The false video detection method of claim 1 , wherein the step of performing the feature extraction procedure on the N×M image frames to obtain the feature map corresponding to the N×M image frames comprises:
performing a data preprocessing operation on the N×M image frames to obtain a first image sequence;
performing a first reshaping operation on the first image sequence to obtain a second image sequence;
inputting the second image sequence into a pre-trained neural network model to obtain a first feature sequence;
performing a second reshaping operation on the first feature sequence to obtain a second feature sequence; and
inputting the second feature sequence into an average pooling layer to obtain the feature map corresponding to the N×M image frames.
3 . The false video detection method of claim 2 , wherein the data pre-processing operation comprises:
performing a normalization operation on each of the N×M image frames such that each pixel of each image frame has a value between 0 and 1; and
performing a data expansion operation on each of normalized N×M image frames to obtain that first image sequence.
4 . The false video detection method of claim 2 , wherein the first reshaping operation comprises performing a dimension decreasing operation on the first image sequence to obtain the second image sequence.
5 . The false video detection method of claim 2 , wherein the pre-trained neural network model comprises one of: a ResNet model, an EffNet model, a VGG16 model, an InceptionV3 model, an Xception model, wherein the pre-trained neural network model performs a feature extraction on the inputted second image sequence to obtain the first feature sequence.
6 . The false video detection method of claim 1 , wherein the time attention model comprises:
a first branch comprising a first average pooling layer, a first convolutional layer, a first function, and a further first convolutional layer;
a second branch comprising a second maximum pooling layer, a second convolutional layer, a second function, a further second convolutional layer; and
a sigmoid function, wherein the feature map is inputted to that first average pooling layer and that second maximum pooling layer of that first branch and that second branch, respectively, and outputted results of the first branch and the second branch are summed and inputted into the sigmoid function to output the concentrated feature.
7 . The false video detection method of claim 6 , wherein the first function and the second function comprise one of: a Leaky ReLU function, a PReLU function, a RReLU function, and a SELU function.
8 . The false video detection method of claim 1 , wherein
a judgment result determined as true indicates that the target video is a real video that does not contain any of faked content, and
a judgment result determined as false indicates that the target video is a false video that contains faked content, wherein the false video was the real video with only the original content before the faked content was added.
9 . A false video detection device, comprising:
a processor, wherein the processor is configured to:
extract N video segments from a target video, wherein each video segment contains M image frames;
perform a feature extraction procedure on the N×M image frames to obtain a feature map corresponding to the N×M image frames;
input the feature map into a time attention model to obtain a concentrated feature;
perform a multiplication operation to the concentrated feature and the feature map to obtain a first enhanced feature map;
perform a summing operation on the first enhanced feature map and the feature map to obtain a second enhanced feature map; and
input the second enhanced feature map into a fully connected layer to obtain a judgment result outputted by the fully connected layer, wherein the judgment result indicates that the target video is true or false.
10 . The false video detection device of claim 9 , wherein the step of performing the feature extraction procedure on the N×M image frames to obtain the feature map corresponding to the N×M image frames comprises:
performing a data preprocessing operation on the N×M image frames to obtain a first image sequence;
performing a first reshaping operation on the first image sequence to obtain a second image sequence;
inputting the second image sequence into a pre-trained neural network model to obtain a first feature sequence;
performing a second reshaping operation on the first feature sequence to obtain a second feature sequence; and
inputting the second feature sequence into an average pooling layer to obtain the feature map corresponding to the N×M image frames.
11 . The false video detection device of claim 10 , wherein the data pre-processing operation comprises:
performing a normalization operation on each of the N×M image frames such that each pixel of each image frame has a value between 0 and 1; and
performing a data expansion operation on each of normalized N×M image frames to obtain that first image sequence.
12 . The false video detection device of claim 10 , wherein the first reshaping operation comprises performing a dimension decreasing operation on the first image sequence to obtain the second image sequence.
13 . The false video detection device of claim 10 , wherein the pre-trained neural network model comprises one of: a ResNet model, an EffNet model, a VGG16 model, an InceptionV3 model, an Xception model, wherein the pre-trained neural network model performs a feature extraction on the inputted second image sequence to obtain the first feature sequence.
14 . The false video detection device of claim 9 , wherein the time attention model comprises:
a first branch comprising a first average pooling layer, a first convolutional layer, a first function, and a further first convolutional layer;
a second branch comprising a second maximum pooling layer, a second convolutional layer, a second function, a further second convolutional layer; and
a sigmoid function, wherein the feature map is inputted to the first average pooling layer and the second maximum pooling layer of the first branch and the second branch, respectively, and outputted results of the first branch and the second branch are summed and inputted into the sigmoid function to output the concentrated feature.
15 . The false video detection device of claim 14 , wherein the first function and the second function comprise one of: a Leaky ReLU function, a PReLU function, a RReLU function, and a SELU function.
16 . The false video detection device of claim 9 , wherein
a judgment result determined as true indicates that the target video is a real video that does not contain any of faked content, and
a judgment result determined as false indicates that the target video is a false video that contains faked content, wherein the false video was the real video with only the original content before the faked content was added.