Summary: ¢¡¤£¦¥¨§©§¥¡¤§£ "! #$ %&!¨#('0)¢21340£657"89£¦4!
£6A@ BC£6)¢ED !¨5GF(HC!I4!¨¥4PQ£RITSU@0¥¨§VW§X§PQ"!
Aishy Amer and Eric Dubois
Montr´eal, Qu´ebec, Canada
In real-time vision (e.g., visual surveillance), fast image seg-
mentation is required. In this paper, a fast unsupervised
artifact-robust segmentation method is presented. Within
this method the whole segmentation process is divided into
simple tasks so that complex operations are avoided.
Herewith, the basic task is the binarization which aims at
finding out significant large object regions. In the paper, two
methods for binarization are proposed: 1) an interference-
invariant thresholding function based on a combination of
local (histogram-based) and global (block-based) decision
criteria. Such a binarization is suitable to segment regions of
low intensity variations; 2) a noise-robust binarization based
on homogeneity tests which implicitly take image noise into