文档介绍：Lecture 8 – Region-based segmentation
Basic Algorithms for Digital Image Analysis 1 Principles of region-based segmentation
Dmitrij Csetverikov Collaborators 2 Region growing and merging
Region growing
Eötvös Loránd University, Budapest, Hungary Region merging
******@
/
3 Split-and-merge
Faculty of Informatics
parison of different techniques
5 Other segmentation methods
Goal of region-based segmentation Region homogeneity
Divide image I into n regions R1, R2, . . . , Rn that are Examples of homogeneity criteria for region R
connected diff. between max and min greyvalues in R is small
homogeneous diff. between any pixel and mean greyvalue in R is small
For this, for any region R ⊂ I define logical homogeneity variance of greyvalues in R is small
criterion P(R) over points of R ⇒ Segmentation depends on
P(R) = TRUE if all pixels in Ri have similar properties properties used
⇒ R is homogeneous measure of similarity between properties
P(R) = FALSE otherwise similarity variation tolerance (threshold)
⇒ R is inhomogeneous
Connectivity in digital images Definition of segmentation
Depends on how many neighbours are considered
connected to pixel
1 Image I is divided into n regions R1, R2, . . . , Rn
8-connectivity: all 8 neighbours n
4-connectivity: only 4 neighbours (vertical, horizontal) 2