Benford’s law to detect
GAN-generated images
What is the probability that the population of any country starts with the digit:
1, 2, 3, 4, 5, 6, 7, 8, or 9?
Is it really equiprobable?
| Country | Population (million) |
|---|---|
| France | 66.99 |
| Belgium | 11.46 |
| ... | ... |
| USA | 328.2 |
Simon Newcomb
Discovered Benford's law in 1881
P(\color{orange}d\color{white})=log_b(1+\frac{1}{\color{orange}d})
Frank Benford
Rediscovered Benford's law in 1938
Pixel value (0 - 255)
Discrete Cosine Transformation (DCT)
Quantized DCT
DCT
Quantization Matrix
K blocks
k-th block 8x8 px
First digit distribution
Results show that GAN-generated images often fail in respecting Benford’s law, thus can be discriminated from natural pictures.
[...], this suggests to possibly embed Benford’s law into GAN loss function in order to obtain even more realistic images.
Thank you !
Data Engineer
WTM Ambassador
Benford's law to detect GAN-generated images
By Tiffany Souterre
