Realistic anatomical images are useful for assessment and improvement of me
dical image quality. The use of synthesized images has the advantage of pro
viding the user with a large number of independent samples, in a controlled
environment. We propose a method to generate medical textures that are ful
ly defined by a set of adjustable parameters and a random number generator,
and which statistical properties are analytically tractable. This method,
called the "clustered lumpy background", is a generalization of the origina
l lumpy background described by Rolland and Barrett (1992). A detailed appl
ication of the method in the case of mammography is presented. It is shown
that the synthesized images are visually very similar and that their first
and second order statistics can be considered as being equivalent. (C) 1999
Optical Society of America.