Printing, photocopying, and scanning processes degrade the image quality of
a document. Statistical models of these degradation processes are crucial
for document image understanding research. Models allow us to predict syste
m performance, conduct controlled experiments to study the breakdown points
of the systems, create large multilingual data sets with groundtruth for t
raining classifiers, design optimal noise removal algorithms, choose values
for the free parameters of the algorithms, and so on. Although research in
document understanding started many decades ago, only two document degrada
tion models have been proposed thus far. Furthermore, no attempts have been
made to statistically validate these models. In this paper, we present a s
tatistical methodology that can be used to validate local degradation model
s. This method is based on a nonparametric, two-sample permutation test. An
other standard statistical device-the power function-is then used to choose
between algorithm variables such as distance functions. Since the validati
on and the power function procedures are independent of the model, they can
be used to validate any other degradation model. A method for comparing an
y two models is also described. It uses p-values associated with the estima
ted models to select the model that is closer to the real world.