Detecting digital human face image manipulations
Systems, devices, methods and instructions are described for detecting GAN generated images. On embodiment involves receiving an images, generating co-occurrence matrices on color channels of the image, generating analysis of the image by using a convolutional neural network trained to analyze image features of the images based on the generated co-occurrence matrices and determining whether the image is a GAN generated image based on the analysis.
1. A method, comprising:
accessing, using one or more processors, an image;
training, using the one or more processors, a convolutional neural network using a set of authentic human face images and a set of digitally manipulated human face images;
generating co-occurrence matrices on pixel values of the image, the generating comprises;
applying a position operator to pixels in the image to determine a defined offsets; and
defining a co-occurrence matrix over the image representing distributions of co-occurring pixel values at the defined offsets;
analyzing the image based on the generated co-occurrence matrices using the trained convolutional neural network, the trained convolutional network trained to detect images generated by a generative adversarial network (GAN); and
classifying the image based on the analysis.
2. The method of claim 1 , wherein the classifying of the image further comprises:
identifying a plurality of pixels within the image that are digitally manipulated by the GAN.
3. The method of claim 2 , wherein the plurality of pixels represents a human face.
4. The method of claim 1 , wherein the set of authentic human face images and the set of digitally manipulated human face images comprise pairs of an original human face image and a corresponding digitally manipulated human face image with at least one varying facial attribute altered by the GAN.
5. The method of claim 1 , wherein the defining of the co-occurrence matrix comprises:
populating the co-occurrence matrices based on co-occurring pixel values at the defined pixel offsets within the image; and
using the populated co-occurrence matrices as inputs to the convolutional neural network.
6. The method of claim 5 , wherein the pixel values comprise grayscale values.
7. The method of claim 1 , wherein the set of authentic human face images and the set of digitally manipulated human face images are joint photographic experts group (JPEG) compressed images.
8. The method of claim 1 , wherein the co-occurrence matrices are generated on pixel values of color channels comprising a red channel, a green channel, and a blue channel.
9. A system comprising:
one or more processors; and
a non-transitory memory storing instructions that, when executed by the one or more processors, configure the system to;
accessing an image;
training a convolutional neural network using a set of authentic human face images and a set of digitally manipulated human face images;
generating co-occurrence matrices on pixel values of the image, the generating comprises:
applying a position operator to pixels in the image to determine a defined offsets; and
defining a co-occurrence matrix over the image representing distributions of co-occurring pixel values at the defined offsets;
analyzing the image based on the generated co-occurrence matrices using the trained convolutional neural network, the trained convolutional network trained to detect images generated by a generative adversarial network (GAN); and
classifying the image based on the analysis.
10. The system of claim 9 , wherein classifying the image further comprises:
identifying a plurality of pixels within the image that are digitally manipulated by the GAN.
11. The system of claim 10 , wherein the plurality of pixels represents a human face.
12. The system of claim 9 , wherein the set of authentic human face images and the set of digitally manipulated human face images comprise pairs of an original human face image and a corresponding digitally manipulated human face image with at least one varying facial attribute altered by the GAN.
13. The system of claim 9 , wherein the defining of the co-occurrence matrix comprises:
populating the co-occurrence matrices based on co-occurring pixel values at the defined pixel offsets within the image; and
using the populated co-occurrence matrices as inputs to the convolutional neural network.
14. The system of claim 13 , wherein the pixel values comprise grayscale values.
15. The system of claim 9 , wherein the set of authentic human face images and the set of digitally manipulated human face images are joint photographic experts group (JPEG) compressed images.
16. The system of claim 9 , wherein the co-occurrence matrices are generated on pixel values of color channels comprising a red channel, a green channel, and a blue channel.
17. A non-transitory computer-readable storage medium, the computer-readable storage medium including instructions that when executed by a computer, cause the computer to perform operations comprising:
accessing an image;
training a convolutional neural network using a set of authentic human face images and a set of digitally manipulated human face images;
generating co-occurrence matrices on pixel values of the image, the generating comprises:
applying a position operator to pixels in the image to determine a defined offsets; and
defining a co-occurrence matrix over the image representing distributions of co-occurring pixel values at the defined offsets;
analyzing the image based on the generated co-occurrence matrices using the trained convolutional neural network, the trained convolutional network trained to detect images generated by a generative adversarial network (GAN); and
classifying the image based on the analysis.
18. The computer-readable storage medium of claim 17 , wherein classifying the image further comprises:
identifying a plurality of pixels within the image that are digitally manipulated by the GAN.