Automated detection of tampered images
View Patent ↗A content analyzer determines whether various types of modification have been made to images. The content analyzer computes JPEG ghosts from the images that are concatenated with the image channels to generate a feature vector. The feature vector is provided as input to a neural network that determines whether the types of modification have been made to the image. The neural network may include a constrained convolution layer and several unconstrained convolution layers. An image fake model may also be applied to determine whether the image was generated using a computer model or algorithm.
1. A method comprising:
receiving an image;
generating one or more JPEG ghosts from the image;
generating a feature vector from the image and the JPEG ghosts;
providing the feature vector as input to a neural network, the neural network including a constrained convolution layer and a plurality of unconstrained convolution layers, wherein the constrained convolution layer includes a plurality of convolution filters, each convolution filter including a central element value and a plurality of other element values, the central element value being a first predetermined value and the other element values summing to a second predetermined value;
generating, by the neural network, an output from the feature vector, the output including an indication of whether a type of modification is present in the image; and
associating metadata with the image indicating whether the type of modification is present in the image based on the output of the neural network.
2. The method of claim 1 , wherein the one or more JPEG ghosts includes a plurality of JPEG ghosts, each JPEG ghost of the plurality computed using a different JPEG compression level.
3. The method of claim 1 , wherein the image includes a plurality of channels and generating the feature vector comprises concatenating pixel values of the plurality of channels with pixel values of the one or more JPEG ghosts.
4. The method of claim 1 , wherein the output from the neural network includes a probability that the type of modification is present in the image, and the metadata is added to the image if the probability exceeds a threshold.
5. The method of claim 1 , wherein the constrained convolution layer is a first layer of the neural network.
6. The method of claim 1 , wherein the output from the neural network identifies whether each of a plurality of types of modification are present in the image and the metadata indicates whether each of the plurality of types of modification are present in the image.
7. The method of claim 6 , wherein the plurality of types of modification include addition of text, having been made into a meme, blurring, object insertion, or requantization.
8. The method of claim 1 , wherein the first predetermined value is negative one and the second predetermined value is one.
9. A non-transitory computer-readable medium storing instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
receiving an image;
generating one or more JPEG ghosts from the image;
generating a feature vector from the image and the JPEG ghosts;
providing the feature vector as input to a neural network, the neural network including a constrained convolution layer and a plurality of unconstrained convolution layers, wherein the constrained convolution layer includes a plurality of convolution filters, each convolution filter including a central element value and a plurality of other element values, the central element value being a first predetermined value and the other element values summing to a second predetermined value;
generating, by the neural network, an output from the feature vector, the output including an indication of whether a type of modification is present in the image; and
associating metadata with the image indicating whether the type of modification is present in the image based on the output of the neural network.
10. The non-transitory computer-readable medium of claim 9 , wherein the one or more JPEG ghosts includes a plurality of JPEG ghosts, each JPEG ghost of the plurality computed using a different JPEG compression level.
11. The non-transitory computer-readable medium of claim 9 , wherein the image includes a plurality of channels and generating the feature vector comprises concatenating pixel values of the plurality of channels with pixel values of the one or more JPEG ghosts.
12. The non-transitory computer-readable medium of claim 9 , wherein the output from the neural network includes a probability that the type of modification is present in the image, and the metadata is added to the image if the probability exceeds a threshold.
13. The non-transitory computer-readable medium of claim 9 , wherein the constrained convolution layer is a first layer of the neural network.
14. The non-transitory computer-readable medium of claim 9 , wherein the output from the neural network identifies whether each of a plurality of types of modification are present in the image and the metadata indicates whether each of the plurality of types of modification are present in the image.
15. The non-transitory computer-readable medium of claim 14 , wherein the plurality of types of modification include addition of text, having been made into a meme, blurring, object insertion, or requantization.
16. The non-transitory computer-readable medium of claim 9 , wherein the first predetermined value is negative one and the second predetermined value is one.