A fixed number of clamped-plate splines (CPS) are placed at a corresponding set of fixed positions. Their direction is chosen stochastically and their intensity (magnitude) is increased to achieve varying levels of perturbations. Linear relationship was assumed between the intensity and the amount of perturbation, i.e. the average number of pixels by which a point is displaced. (For more details on the actual implementation, ask Vlad. Carole says that it is aggregation of warps that increases the perturbation. So, added displacement is only known vaguely)
The expected pitfalls are as follows: