Slide 1

Texture Featuring Based on
Watershed
Karsten Rodenacker, Uta Jütting
GSF-Institut für Biomathematik und Biometrie
Neuherberg
11.05.98


Slide 2

Content
Introduction
 Some remarks about watershed
 Textural featuring
 Examples
 Summary



Slide 3

Introduction


Textures


Slide 4

Some remarks about watershed


Definition
 Given

an image and a local maximum at point
m. Consider all monotonous decreasing paths
w(m,l) starting from m and ending in l.
 The set of all points l is a connected region.
 The set of all such regions is a partition of the
input image, the order corresponds with the
number of local maximums.
 The borders of the regions are the watersheds


Slide 5

Some remarks about watershed


Definition (continued)
 The

inversion of the above definition
concerning maximum by minimum and
decreasing by increasing delivers a second type
of watershed.


Slide 6

Some remarks about watershed


Explanation
 1-dimensional


Slide 7

Some remarks about watershed


Explanation
 2-dimensional


Slide 8

Some remarks about watershed


Explanation
 2-dimensional


Slide 9

Some remarks about watershed


Result of transformation
 Input

image (original, smoothed 3x3)


Slide 10

Some remarks about watershed


Result of transformation
 Watersheds

(labelled regions)


Slide 11

Some remarks about watershed


Result of transformation
 Watersheds


Slide 12

Some remarks about watershed


Result of transformation
 Lower,

upper ricefield


Slide 13

Some remarks about watershed


Result of transformation
 Half

height segmentation HU (black)/HL (white)


Slide 14

Textural featuring


global
 Statistical

estimators (moments) from:

– half height segmentations and original
– topological gradient
 Derived

parameters

– stereological (volume densities)
– densitometric
– morphometric


Slide 15

Textural featuring


global, e.g.
V VU 
P AU 

HUA

volume density

A
HUA

mean particle area

HUNO

E  HUNO  HLNO
RDD 

H U M 1  H LM 1
M1

Euler number
Rel. density difference


Slide 16

Textural featuring


heuristical
 neighborhood

related of connected regions
(graph theoretical approach)
– connectivity
– distances

 neighborhood AND

– pattern matching
 ...

intensity related


Slide 17

Examples


global texture features:
 VVU

Volume density
 RDD Relative density difference


Slide 18

Examples


Projects
 Differentiation

of hormone status of breast
carcinoma patients
VVU, RDD
 Differentiation of osteoblasts under different
growth conditions (Osteoporosis)
RHUA
 Differentiation of neuroendocrine lung tumours
HUCV


Slide 19


Slide 20

Examples


global texture features: RDD


Slide 21

Examples


Slide 22

Examples


Projects


Slide 23

Summary
+ Topology by watershed
 Taxonomy

of transformation results

+ Segmentation
+ Parameter free featuring
- Noise sensitivity of watershed
- Difficulties of interpretation for natural
images (reflected light, shadows, ...)