Color Image Segmentation based on Automatic Morphological - - PowerPoint PPT Presentation

▶
color image segmentation
SMART_READER_LITE
LIVE PREVIEW

Color Image Segmentation based on Automatic Morphological - - PowerPoint PPT Presentation

Jul 08, 2023 228 likes •472 views

Color Image Segmentation based on Automatic Morphological Clustering T. Graud, P.-Y. Strub, J. Darbon thierry.geraud @ lrde.epita.fr EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 1 Outline

slide-1
SLIDE 1

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 1

Color Image Segmentation

based on

Automatic Morphological Clustering

  • T. Géraud, P.-Y. Strub, J. Darbon

thierry.geraud@lrde.epita.fr

slide-2
SLIDE 2

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 2

Outline

  • Introduction

– about statistical classification – about watershed algorithm – problem statement

  • Morphological classification

– state of the art – description of proposed approach – commented results

  • Conclusion
slide-3
SLIDE 3

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 3

A classical statistical and non-contextual classification scheme

  • Transform observations into feature vectors

– for a pixel, a feature can be a color component, a local variance… – difficulty: find a relevant feature space

  • In feature space

– assign / learn a parametric model for each class – then run a classifier Remark: the probability density function of a class in the feature space can be estimated from few samples; e.g., convolve the samples with a Gaussian kernel

slide-4
SLIDE 4

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 4

About watershed algorithm

  • Key features

it applies on n -D images the algorithm divides the input image into regions (basins)

  • ne local minimum leads to one surrounding basin

a 1-pixel thick component (watershed) separates every basins basin boundaries are located on image crest values

  • Connected version of the algorithm

the watershed itself is suppressed

  • ther properties are maintained

as output image we get a partition

  • A reliable segmentation tool

“Scale-Space Segmentation of Color Images Using Watersheds and Fuzzy Region Merging,” by Makrogiannis et al., ICIP 2001

slide-5
SLIDE 5

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 5

A classical morphological segmentation method

morphological gradient

(high values correspond to object contours)

morphological closing

(the number of local minima is reduced)

morphological watershed algorithm

(the watershed is located on object contours)

slide-6
SLIDE 6

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 6

Problem statement

  • Color images

– feature space is (at least) 3-dimensional – in such a space, clusters have low-density – cluster cardinalities are very heterogeneous – many artifacts appear

due to:

  • storage compression
  • color gradations
  • specular surface of objects
  • Statistical models

are they relevant?

r e d g r e e n

log(P/RG for all B(H))

slide-7
SLIDE 7

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 7

Morphological classification of color images (state of the art)

  • Basic idea:

RGB image compute histogram = 3D image morphological cluster identification

Postaire et al., “Cluster Analysis by Binary Morphology”, PAMI 15(2). Zang et al., “Convexity Dependent Morphological transformations for Mode Detection in Cluster Analysis,” Pattern Recognition 27(1). Park et al., “Color Image Segmentation Based on 3D Clustering: Morphological Approach,” Pattern Recognition 31(8).

  • Drawbacks:

clusters should be prominent and well-contrasted

  • nly cluster cores are segmented; so, how to handle color outliers?
slide-8
SLIDE 8

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 8

Morphological classification

presented here

  • From a color image:

express data in feature space

for instance, a 3-D RGB histogram

consider data as a n-D image regularize data run a morphological partitioning

  • Originality:

use of the watershed algorithm as a classifier

slide-9
SLIDE 9

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 9

Method details

Step Description Rationale ………………………………………………………………...

1 data computation in feature get a grey-level image H where space, log transform, and inversion clusters have dark values 2 Gaussian filtering regularize (while suppressing many local minima) 3 closing plus cutting low values suppress extra local minima 4 connected watershed algorithm get a partition W of feature space 5 apply a segmentation process...

slide-10
SLIDE 10

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 10

Method properties

  • Applying an increasing function f to feature space values

(densities): Hbis(c) = f (H(c)) => Wbis(c) = W(c)

  • Applying a rigid transform T to features:

Hbis(c’) = H(T(c)) => Wbis(c’) = W(T(c))

  • Applying a scaling factor α to a given feature:

Hbis(c1, c’2) = H(c1,α c2) => Wbis(c1, c’2) = W(c1,α c2)

slide-11
SLIDE 11

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 11

Some segmentation approaches

(step 5)

  • Using directly feature space partitioning:

segmentation = non contextual labeling but a feature can be contextual (e.g., a local variance)

  • Considering that we can learn from feature space classes…

for example, perform a Bayesian labeling: estimate Mahalanobis distances from basins run a Markovian relaxation in image domain

slide-12
SLIDE 12

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 12

Projections on the RG plane of 3D data:

Segmentation results

(on peppers image)

log( P/RG(H) ) result of step 3

  • uzo

= input of the watershed algorithm

slide-13
SLIDE 13

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 13

result of step 3 basin boundary local minima

slide-14
SLIDE 14

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 14

classes log( P/RG(H) )

slide-15
SLIDE 15

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 15

Non-contextual labeling

slide-16
SLIDE 16

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 16

Markovian labeling

slide-17
SLIDE 17

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 17

log( P/RG(H) ) classes

Other results

  • riginal

Markovian labeling

slide-18
SLIDE 18

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 18

log( P/RG(H) ) clusters

lena jbeanc tiffany

slide-19
SLIDE 19

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 19

What about results from extreme data?

(oops… so many clusters! It should be a...)

log( P/RG(H) )

slide-20
SLIDE 20

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 20

...Kandinsky

slide-21
SLIDE 21

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 21

log( P/RG(H) ) classes

slide-22
SLIDE 22

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 22

part of

  • riginal

image non- contextual labeling

slide-23
SLIDE 23

EPITA Research and Development Laboratory, France / ICIP, Thessaloniki, October 2001 23

  • Original use of the connected watershed algorithm:

leads to an automatic classification method is applied to color image segmentation provides rather good and robust results

http://www.lrde.epita.fr/download

  • But:

needs to be refined by merging (to improve the segmentation) and/or splitting classes (to serve as an halftoning method) cannot separate two clusters when they closely mix is memory consuming (3D feature space)

Conclusion