MIR: An Approach to Robust Clustering
- Application to Range Image
Segmentation
K. Koester, M. Spann
IEEE Trans. PAMI, Vol 22(5), May 2000, pp 430-444
Abstract
This paper describes an unsupervised region growing technique based on
a novel robust statistical test. The merging decision is derived from
the
mutual inlier ratio (MIR) of adjacent regions. This ratio is computed
using
robust regression techniques [Rousseeuw, 1987] and a novel method to
estimate
the robust scale of the Gaussian distribution. A discrimination value
to
recognize identical Gaussian distributions with the MIR is derived
theoretically
as a function of the sizes of the compared sets. The method to test
distributions
is compared with the established Kolmogorov-Smirnov test and
implemented
into a segmentation algorithm for planar range images. The iterative
growing
technique is evaluated using the detailed framework for range image
segmentation
comparison of Hoover et. al. [Hoover, 1996] involving 60 real range
images.
Download gzipped Postscript (1091K)
Project
achievements
Back
to papers and internal reports
