Publication View

Layers-Based Image Segmentation Incorporating Motion Estimation with Static Segmentation (2007)

Abstract
This paper addresses the problem of motion segmentation using a multilayer representation. At first, the coarse optic flow is estimated using the robust Simultaneous-Over-Relaxation (SOR) algorithm, meanwhile the intensity segmentation is performed by a watershed algorithm. The image is divided into nonoverlapping rectangular tiled regions, within those subregions affine motion models are fitted from the optic flow. Using the ISODATA clustering algorithm, those affine models are grouped into a small number of motion models. Finally, based on registration errors by image warping on each watershed segment, the support maps of multilayers are obtained. Experimental results from real images are given to show the efficiency of our method.

Publication details
Download http://citeseerx.ist.psu.edu/viewdoc/summary?doi=?doi=10.1.1.30.1485
Source http://www5.informatik.uni-erlangen.de/HTML/English/Persons/Guests/yuhuang/irish.ps.gz
Contributors CiteSeerX
Repository CiteSeerX - Scientific Literature Digital Library and Search Engine (United States)
Keywords Optic flow, ISODATA algorithm, watershed, robust estimation, static segmentation
Type text
Language English
Relation 10.1.1.40.5046, 10.1.1.50.8541, 10.1.1.27.2370, 10.1.1.54.7470