Texture analysisTexture analysisTeam 5Team 5

Alexandra Bulgaru Alexandra Bulgaru Justyna Jastrzebska Justyna Jastrzebska

Ulrich Leischner Ulrich Leischner Vjekoslav Levacic Vjekoslav Levacic

Güray TonguçGüray Tonguç

ContentsContents

Project goalProject goal Definition of textureDefinition of texture Features used in texture analysisFeatures used in texture analysis Example of application for texture Example of application for texture

based image querybased image query ResultsResults ConclusionsConclusions

Project goalProject goal

Defining a set of features which would Defining a set of features which would help in identifying the textures in the help in identifying the textures in the imageimage

Examining the relation between features Examining the relation between features and the texturesand the textures

Defining a simple set of features to Defining a simple set of features to identify similar textures in texture identify similar textures in texture databasedatabase

Possibility of using texture classification Possibility of using texture classification and segmentation in later applicationsand segmentation in later applications

Definition of a textureDefinition of a texture Texture Texture is used to describe two dimensional is used to describe two dimensional

arrays of variationarrays of variation.. The elementsThe elements and rules of spacing orand rules of spacing or

arrangement arrangement in texture in texture may be arbitrarilymay be arbitrarily manipulated, providedmanipulated, provided a characteristic a characteristic repetitiveness remains.repetitiveness remains.

Features used in texture Features used in texture analysisanalysis

Problem of feature selection depends on:Problem of feature selection depends on: Type of application (medical, aerial, etc.)Type of application (medical, aerial, etc.) Need of invariances (rotational, shifting, scaling, Need of invariances (rotational, shifting, scaling,

lightning, etc.)lightning, etc.)

Examples of features we used:Examples of features we used: StatisticalStatistical (for example derived from co- (for example derived from co-

occurrence matrix like entropy, contrast, occurrence matrix like entropy, contrast, correlation)correlation)

High levelHigh level (derived from the watershed algorithm) (derived from the watershed algorithm) Frequency domainFrequency domain (energy bands) (energy bands)

Co – occurrence matrixCo – occurrence matrix

Original image

00 11 22 33

00 22 22 11 00

11 00 22 00 00

22 00 00 33 11

33 00 00 00 11

neighbour pixel value

Ref

eren

ce p

ixe

l va

lue:

Contrast- feature derived from the Co- occurrence matrix

Calculation of the Calculation of the feature valuefeature value

0.160.16 0.080.08 0.04 0.04 00

0.080.08 0.16 0.16 00 00

0.040.04 00 00.24.24 0.040.04

00 00 0.040.04 0.080.08

00 11 44 99

11 00 11 44

44 11 00 11

99 44 11 00

Contrast- feature derived form the Co -

occurrence matrix

Original image

x =

00 0.00.088

0.10.166

00

0.00.088

00 00 00

0.10.166

00 00 0.00.044

00 00 0.00.044

00Standardized symmetric matrix

Sum of all cells

=

0.586

Statistical features - Statistical features - EntropyEntropy

)ln( iii

ppEntropy pixels ofnumber total

i valueintensity with pixels ofnumber ip

Original image

Entropy

Watershed segmentationWatershed segmentation Average area of componentsAverage area of components Number of components in specific regionNumber of components in specific region Ratio between circumference and components Ratio between circumference and components

numbernumber

Watershed from original image

Watershed form the median filtered image (smoothing of noise

to avoid oversegmentation)

Original image

Watershed analysis – Watershed analysis – Average area of Average area of

componentscomponents

With a big filter size: Better features inside but borders are imprecise

Watershed form the median

filtered image

Original image

Frequency domain Frequency domain featurefeature

Low to high spectrum Low spectrum Periodicity of image

Example of application for Example of application for texture based image querytexture based image query

Concept of workConcept of work

f1

We wanted to represent each texture as a feature We wanted to represent each texture as a feature vectorvector

Each texture Fn, where n is number of textures Each texture Fn, where n is number of textures in the database, will be noted as unique classin the database, will be noted as unique class

f2 fnE … = F1

Which classifier to use?Which classifier to use?

SVMSVM Can be used if multiple classifiers are used Can be used if multiple classifiers are used

but there are problems with small number but there are problems with small number of training vectors and large number of of training vectors and large number of classesclasses

Our solutionOur solution Definition of a measure – Euclidean distanceDefinition of a measure – Euclidean distance Simple comparing the length between input Simple comparing the length between input

feature vector with those in database and feature vector with those in database and taking the closesttaking the closest

Problems to addressProblems to address

In large database of textures how to In large database of textures how to compare the feature vectors fastcompare the feature vectors fast

Using the features which are invariant Using the features which are invariant to different transformationsto different transformations

How to include more sophisticated How to include more sophisticated measure which will favor selecting the measure which will favor selecting the feature vector of the texture in a feature vector of the texture in a database which resembles most to the database which resembles most to the input imageinput image