Identification of neural-network-generated fake images
A computer that identifies a fake image is described. During operation, the computer receives an image. Then, the computer performs analysis on the image to determine a signature that includes multiple features. Based at least in part in the determined signature, the computer classifies the image as having a first signature associated with the fake image or as having a second signature associated with a real image, where the first signature corresponds to a finite resolution of a neural network that generated the fake image, a finite number of parameters in the neural network that generated the fake image, or both. For example, the finite resolution may correspond to floating point operations in the neural network. Moreover, in response to the classification, the computer may perform a remedial action, such as providing a warning or a recommendation, or performing filtering.
1. A computer system, comprising:
a computation device;
memory configured to store program instructions, wherein, when executed by the computation device, the program instructions cause the computer system to perform one or more operations comprising:
receiving an image;
classifying, based at least in part on multiple features associated with the image, the image as having a first signature associated with a fake image or as having a second signature associated with a real image, wherein the first signature corresponds to a finite resolution of a neural network that generated the fake image, a finite number of parameters in the neural network that generated the fake image, or both; and
selectively performing a remedial action based at least in part on the classification.
2. The computer system of claim 1 , wherein the finite resolution corresponds to floating point operations in the neural network.
3. The computer system of claim 1 , wherein the first signature corresponds to differences between the image and, given locations of one or more light sources and one or more objects in the image, predictions of a physics-based rendering technique.
4. The computer system of claim 1 , wherein the computer system is configured to implement a second neural network that is configured to perform the classification.
5. The computer system of claim 4 , wherein the second neural network comprises a generative adversarial network (GAN).
6. The computer system of claim 4 , wherein a magnification of the image during the classification may be based at least in part on how long the neural network was trained.
7. The computer system of claim 1 , wherein the image comprises a video with a sequence of images.
8. The computer system of claim 7 , wherein the first signature is associated with a noise property of a shadow region located proximate to and at a former position of a moving object in the sequence of images in the video.
9. The computer system of claim 8 , wherein the noise property comprises a speckled pattern of greyscale values in the shadow region in a given image in the sequence of images.
10. The computer system of claim 1 , wherein the one or more operations comprise performing analysis on the image to determine the multiple features; and
wherein the analysis comprises:
a feature extraction technique that determines a group of basis features; and
selecting a subset of the basis features, which spans a space that includes the first signature and the second signature.
11. The computer system of claim 1 , wherein the classification is performed using a classifier that was trained using a supervised learning technique and a training dataset with predefined signatures; and
wherein the predefined signatures comprise at least the first signature and the second signature.
12. The computer system of claim 1 , wherein the classification is performed using a classifier that was trained using images that were classified using an unsupervised learning technique.
13. The computer system of claim 1 , wherein the remedial action comprises one of:
providing a warning associated with the image; providing a recommendation associated with the image; or filtering at least a portion of the content in the image.
14. A non-transitory computer-readable storage medium for use in conjunction with a computer system, the computer-readable storage medium configured to store program instructions that, when executed by the computer system, causes the computer system to perform one or more operations comprising:
receiving an image;
classifying, based at least in part on multiple features associated with the image, the image as having a first signature associated with a fake image or as having a second signature associated with a real image, wherein the first signature corresponds to a finite resolution of a neural network that generated the fake image, a finite number of parameters in the neural network that generated the fake image, or both; and
selectively performing a remedial action based at least in part on the classification.
15. The non-transitory computer-readable storage medium of claim 14 , wherein the finite resolution corresponds to floating point operations in the neural network.
16. The non-transitory computer-readable storage medium of claim 14 , wherein the first signature corresponds to differences between the image and, given locations of one or more light sources and one or more objects in the image, predictions of a physics-based rendering technique.
17. The non-transitory computer-readable storage medium of claim 14 , wherein the computer system is configured to implement a second neural network that is configured to perform the classification.
18. The non-transitory computer-readable storage medium of claim 14 , wherein the image comprises a video with a sequence of images; and
wherein the first signature is associated with a noise property of a shadow region located proximate to and at a former position of a moving object in the sequence of images in the video.
19. The non-transitory computer-readable storage medium of claim 14 , wherein the one or more operations comprise performing analysis on the image to determine the multiple features; and
wherein the analysis comprises:
a feature extraction technique that determines a group of basis features; and
selecting a subset of the basis features, which spans a space that includes the first signature and the second signature.
20. A method for identifying a fake image, wherein the method comprises:
by a computer system:
receiving an image;
classifying, based at least in part on multiple features associated with the image, the image as having a first signature associated with a fake image or as having a second signature associated with a real image, wherein the first signature corresponds to a finite resolution of a neural network that generated the fake image, a finite number of parameters in the neural network that generated the fake image, or both; and
selectively performing a remedial action based at least in part on the classification.