Xiang Yang(1) and Peter Meer(2)
(1)Dept. of Mechanical and Aerospace Engineering
(2) Dept. of Electrical and Computer Engineering
Rutgers University, NJ 08854, USA
(AN EARLIER VERSION.)
In this paper we present a method to robustly estimate
multiple inlier structures with different scales in the presence of noise.
The estimation is done iteratively with the objective function transformed
into a higher dimensional linear space by carrier vectors.
An initial set consisting of
a small number of points that has the minimum
sum of Mahalanobis distances is detected from the
trials based on elemental subsets.
The region of interest is defined by applying an expansion criteria to
an increasing sequence of sets of points
which begins with the initial set and increases
until the set cannot expand.
The largest expansion in this region gives the scale estimate.
The original mean shift is applied to all remaining
input points to re-estimate the structure.
After all data are processed,
the segmented structures are sorted by strengths
with the strongest inlier structures at the front.
Several synthetic and real examples are presented to
illustrate every aspect of the algorithm.
This method is easy to implement and its limitations of
robustness are clearly stated.
arxiv.org, Can be seen in the archive too as 1609.06371.
Submitted September 20, 2016.
Return to Research: Robust Analysis of Visual Data