About one year ago I did a project on Bill-Board Cloud (http://blog.csdn.net/saintony/archive/2010/04/28/5541052.aspx), and the refered paper is [Lacawell06JGT] "Stochastic Billboard Clouds for Interactive Foliage Rendering".
I didn't understand the 2nd step of its original algorithm: "Perturb the vertices of Ts by independent random distances". I thought it is totally unnecessary. But after reviewing this paper again one year later, I understand why the author did this: to introduce randomness.
According to CLRS (some place in the first several chapters, the employment optimization problem), the order of all candidates should be perturbed randomly before going ahead - so, it means sometimes inputs from natural world may be not distributed randomly enough (shocking, actually) so doing this perturb can enforce this randomness, and then to have a better average algorithm output.
So is the same as [Lacawell06JGT]...
See? CLRS works again!