Traditionally, the goal of image segmentation has been to produce a si
ngle partition of an image. This partition is compared to some 'ground
truth', or human approved partition, to evaluate the performance of t
he algorithm. This paper utilizes a framework for considering a range
of possible partitions of the image to compute a probability distribut
ion on the space of possible partitions of the image. This is an impor
tant distinction from the traditional model of segmentation, and has m
any implications in the integration of segmentation and recognition re
search. The probabilistic framework that enables us to return a confid
ence measure on each result also allows us to discard from considerati
on entire classes of results due to their low cumulative probability.
The distributions thus returned may be passed to higher-level algorith
ms to better enable them to interpret the segmentation results. Severa
l experimental results are presented using Markov random fields as tex
ture models to generate distributions of segments and segmentations on
textured images. Both simple homogeneous images and natural scenes ar
e presented. (C) 1997 Elsevier Science B.V.