Proposed by Cox, Killian, Lathan and Shamoon.

This algorithm takes the image and breaks it up into its Discrete Cosine Transform.

Leaving the first discrete co-efficient aside, use the first 1000 (this can vary) co-efficients taken along the diagonal.

For each pixel pixel value P, we calculate P = P + A * X;

where, A = real number whose value is normally 0.1; X = real number which is part of the pseudo-random number sequence (this has a normal distribution).

The new pixel values are calculated using the inverse of the DCT.

Detection is done by regenerating the pseudo-random numbers and comparing the statistical property of the new image with the original one.