Scenes are made up of objects and the variability in range pixels
results from variations in: (i) objects' shapes and (ii) their
placements in a scene. For objects with known shapes, we assume a
stochastic model on their placements to derive a physical
probability model on range pixel. This model leads to univariate
and multivariate probability densities on pixel values. For
unknown objects present in a scene, we adopt a two-parameter
family of probability densities, introduced in
\cite{grenander-srivastava-clutter} and called  {\bf Bessel K
forms}, to model the range pixels and their extracted features.
Results based on real and simulated range images of forest scenes
are presented.