Method and apparatus for classifying fake images
An apparatus for classifying fake images according to an embodiment of the present disclosure includes an artifact remover configured to receive an input image to generate an artifact-removed image from which artifacts are removed, an artifact image generator configured to generate an artifact image by using a difference between the input image and the artifact removal image, and a determiner configured to determine whether the artifact image is a real image or a fake image by receiving the artifact image.
1. An apparatus for classifying fake images comprising:
an artifact remover configured to receive an input image to generate an artifact-removed image from which artifacts are removed;
an artifact image generator configured to generate an artifact image by using a difference between the input image and the artifact-removed image; and
a determiner configured to determine whether the artifact image is a real image or a fake image by receiving the artifact image,
wherein the artifact remover comprises:
an artifact-removed image generator configured to generate the artifact-removed image from the input image by using an artificial neural network trained to remove artifacts of the input image; and
an artifact-removed image determiner configured to be trained to determine whether the artifact-removed image is a real image or a fake image, and
wherein the artifact-removed image generator is further configured to be trained based on:
an adversarial loss function defined so that an artifact-removed image of a real input image and an artifact-removed image of a fake input image are similar, and
a normalized loss function defined so that the artifact-removed image of the real input image is similar to the real input image and the artifact-removed image of the fake input image is made close to zero.
2. The apparatus of claim 1 , wherein the
artificial neural network is trained to remove the artifacts of the input image in a frequency domain.
3. The apparatus of claim 2 , wherein the artifact-removed image generator is further configured to:
convert the input image into a frequency domain to generate a converted image,
remove artifacts from the converted image by using the artificial neural network, and
generate an artifact-removed image by converting the artifact-removed converted image into an image domain.
4. The apparatus of claim 2 , wherein the artifact-removed image generator and the artifact-removed image determiner constitute a generative adversarial network (GAN).
5. The apparatus of claim 1 , wherein the determiner comprises:
a first determiner configured to determine whether an image is a real image or a fake image based on the artifact image; and
a second determiner configured to convert the artifact image into a frequency domain and then determines whether the converted image is a real image or a fake image.
6. A method for classifying fake images comprising:
removing artifacts by receiving an input image to generate an artifact-removed image from which the artifacts are removed;
generating an artifact image by using a difference between the input image and the artifact-removed image; and
determining whether the artifact image is a real image or a fake image by receiving the artifact image,
wherein the removing of the artifacts comprises:
generating the artifact-removed image from the input image by using an artificial neural network trained to remove artifacts of the input image; and
determining whether the artifact-removed image is a real image or a fake image, and
wherein the artificial neural network is trained based on:
an adversarial loss function defined so that an artifact-removed image of a real input image and an artifact-removed image of a fake input image are similar, and
a normalized loss function defined so that the artifact-removed image of the real input image is similar to the real input image and the artifact-removed image of the fake input image is made close to zero.
7. The method of claim 6 , wherein the
artificial neural network is trained to remove the artifacts of the input image in a frequency domain.
8. The method of claim 7 , wherein the generating of the artifact-removed image comprises:
generating a converted image by converting the input image into a frequency domain;
removing artifacts from the converted image by using the artificial neural network, and
generating the artifact-removed image by converting the artifact-removed converted image into an image domain.
9. The method of claim 7 , wherein the generating of the artifact-removed image and the determining of the artifact-removed image constitute a generative adversarial network (GAN).
10. The method of claim 6 , wherein the determining comprises:
determining whether an image is a real image or a fake image based on the artifact image; and
converting the artifact image into a frequency domain and then determining whether the converted image is a real image or a fake image.