ch 10 image segmentation

21
CSE 803 Fall 2008 Stockma n 1 Ch 10 Image Segmentation Ideally, partition an image into regions corresponding to real world objects.

Upload: aspen

Post on 09-Feb-2016

82 views

Category:

Documents


0 download

DESCRIPTION

Ch 10 Image Segmentation. Ideally, partition an image into regions corresponding to real world objects. Goals of segmentation. Segments formed by K-means. Segmentation attempted via contour/boundary detection. Clustering versus region-growing. Clustering versus region-growing. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Ch 10 Image Segmentation

CSE 803 Fall 2008 Stockman 1

Ch 10 Image Segmentation

Ideally, partition an image into regions corresponding to

real world objects.

Page 2: Ch 10 Image Segmentation

CSE 803 Fall 2008 Stockman 2

Goals of segmentation

Page 3: Ch 10 Image Segmentation

CSE 803 Fall 2008 Stockman 3

Segments formed by K-means

Page 4: Ch 10 Image Segmentation

CSE 803 Fall 2008 Stockman 4

Segmentation attempted via contour/boundary detection

Page 5: Ch 10 Image Segmentation

CSE 803 Fall 2008 Stockman 5

Clustering versus region-growing

Page 6: Ch 10 Image Segmentation

CSE 803 Fall 2008 Stockman 6

Clustering versus region-growing

Page 7: Ch 10 Image Segmentation

CSE 803 Fall 2008 Stockman 7

K-means clustering as before: vectors can contain color+texture

Page 8: Ch 10 Image Segmentation

CSE 803 Fall 2008 Stockman 8

K-means

Page 9: Ch 10 Image Segmentation

CSE 803 Fall 2008 Stockman 9

Histograms can show modes

Page 10: Ch 10 Image Segmentation

CSE 803 Fall 2008 Stockman 10

Otsu’s method assumes K=2. It searches for the threshold t that optimizes the intra class variance.

Page 11: Ch 10 Image Segmentation

CSE 803 Fall 2008 Stockman 11

Ohlander bifurcated the histogram recursively

Page 12: Ch 10 Image Segmentation

CSE 803 Fall 2008 Stockman 12

Recursive histogram-controlled segmentation

Page 13: Ch 10 Image Segmentation

CSE 803 Fall 2008 Stockman 13

URL’s of other work Segmentation lecture

http://robots.stanford.edu/cs223b/index.html 

Tutorial on graph cut method and assessment of segmentation http://www.cis.upenn.edu/~jshi/GraphTutorial/

Page 14: Ch 10 Image Segmentation

CSE 803 Fall 2008 Stockman 14

Segmentation via region-growing (aggregation)

Pixels, or patches, at the lowest level are combined

when similar in a hierarchical fashion

Page 15: Ch 10 Image Segmentation

CSE 803 Fall 2008 Stockman 15

Decision: combine neighbors?

Neighboring pixel or region

Page 16: Ch 10 Image Segmentation

CSE 803 Fall 2008 Stockman 16

Aggregation decision

Page 17: Ch 10 Image Segmentation

CSE 803 Fall 2008 Stockman 17

Representation of regions

Page 18: Ch 10 Image Segmentation

CSE 803 Fall 2008 Stockman 18

Chain codes for boundaries

Page 19: Ch 10 Image Segmentation

CSE 803 Fall 2008 Stockman 19

Quad trees divide into quadrants

M=mixed; E=empty; F=full

Page 20: Ch 10 Image Segmentation

CSE 803 Fall 2008 Stockman 20

Can segment 3D images also Oct trees subdivide into 8 octants Same coding: M, E, F used Software available for doing 3D

image processing and differential equations using octree representation.

Can achieve large compression factor.

Page 21: Ch 10 Image Segmentation

CSE 803 Fall 2008 Stockman 21

More URLs of other work Mean shift description

http://cmp.felk.cvut.cz/cmp/courses/ZS1/slidy/meanShiftSeg.pdf