A multiscale regularity measure as a geometric criterion for image segmentation.

Autor: Vasselle, Bruno, Giraudon, Gérard
Zdroj: Machine Vision & Applications; Dec1994, Vol. 7 Issue 4, p229-236, 8p
Abstrakt: Recent results from human vision experiments show that lines of low fractal dimension are highly capable of evoking indification with nameable objects. In other words, regular lines are recognized in human vision as object edges. In this paper, a regularity measure of discrete line geometry is presented. This quantitative measure based on a ratio between lines of varying lengths is analyzed in the framework of brownian motion theory. The measure on a given scale is always computed from the maximum precision image, so that no subresolution assumption is introduced. A choice of scale determines the quantity of global information versus local information one wants to measure. We show how this quantitative measure leads to relevant shape information. To illustrate this, an example of an image segmentation application is realized. The segmentation based essentially on geometry criteria, uses a region-growing process. The process depends on a single parameter that can be fixed in a natural way, comparing contour regularity to a geometric model regularity. We present experimental results performed on real-scene images, including indoor and outdoor images. [ABSTRACT FROM AUTHOR]
Databáze: Complementary Index