Ch 10 Image Segmentation Ideally partition an image

Ch 10 Image Segmentation Ideally, partition an image into regions corresponding to real world objects. CSE 803 Fall 2012 1

Goals of segmentation CSE 803 Fall 2012 2

Segments formed by K-means CSE 803 Fall 2012 3

Segmentation attempted via contour/boundary detection CSE 803 Fall 2012 4

Clustering versus region-growing CSE 803 Fall 2012 5

Clustering versus region-growing CSE 803 Fall 2012 6

K-means clustering as before: vectors can contain color+texture CSE 803 Fall 2012 7

K-means CSE 803 Fall 2012 8

Histograms can show modes CSE 803 Fall 2012 9

Otsu’s method assumes K=2. It searches for the threshold t that optimizes the intra class variance. CSE 803 Fall 2012 10

Ohlander bifurcated the histogram recursively CSE 803 Fall 2012 11

Recursive histogram-controlled segmentation CSE 803 Fall 2012 12

URL’s of other work n n Segmentation lecture http: //robots. stanford. edu/cs 223 b/index. html Tutorial on graph cut method and assessment of segmentation http: //www. cis. upenn. edu/~jshi/Graph. T utorial/ CSE 803 Fall 2012 13

Segmentation via regiongrowing (aggregation) Pixels, or patches, at the lowest level are combined when similar in a hierarchical fashion CSE 803 Fall 2012 14

Decision: combine neighbors? Neighboring pixel or region CSE 803 Fall 2012 15

Aggregation decision CSE 803 Fall 2012 16

Representation of regions CSE 803 Fall 2012 17

Chain codes for boundaries CSE 803 Fall 2012 18

Quad trees divide into quadrants M=mixed; E=empty; F=full CSE 803 Fall 2012 19

Can segment 3 D images also n n Oct trees subdivide into 8 octants Same coding: M, E, F used Software available for doing 3 D image processing and differential equations using octree representation. Can achieve large compression factor. CSE 803 Fall 2012 20

More URLs of other work n Mean shift description http: //cmp. felk. cvut. cz/cmp/courses/ZS 1/slidy/mean. Shift. Seg. pdf CSE 803 Fall 2012 21