Recognising fake images using frequency analysis

EurekAlert  July 16, 2020
To date, deep-fake images have been analysed using complex statistical methods. Researchers in Germany converted the images into the frequency domain using the discrete cosine transform and express the image as the sum of many different cosine functions. Natural images consist mainly of low-frequency functions. The analysis has shown that images generated by GANs (Generative Adversarial Networks) exhibit artefacts in the high-frequency range. The researchers’ experiments showed that these artefacts do not only occur in GAN generated images. They are a structural problem of all deep learning algorithms. Frequency analysis is therefore an effective way to automatically recognise computer-generated images. The researchers make their code freely available online so that other groups can reproduce their results. Their paper was presented at a recent machine learning conference…read more.

Posted in Imaging technology and tagged , .

Leave a Reply