IMAGE RE-SEGMENTATION A new approach applied to Urban Imagery Thales Sehn Korting Leila Maria Garcia...

Preview:

Citation preview

IMAGE RE-SEGMENTATIONA new approach applied to Urban Imagery

Thales Sehn KortingLeila Maria Garcia Fonseca

Luciano Vieira DutraFelipe Castro da Silva

Introduction

• Image segmentation is the identification of homogeneous regions in the image

Traditional approaches

• Few versions available for end users

• Algorithms don’t consider the application context– Urban– Agriculture– etc

Urban Segmentation

Objective

• Development of a re-segmentation system, based on rectangular shapes

• Re-Segmentation– Rearrangement of a polygon set, merging

some elements to generate objects with particular characteristics, applied to a specific context

Re-Segmentation Approach

Band 1

Band 2

Band ...Band ...Band ...Band n

Input

Output

Ove

r-S

egm

enta

tion

Segmentation Based on Graph

• Region Adjacency Graph

Re-Segmentation Diagram

Finding Rectangles

Pi

x

y

AV

ANG(Pi)Ri

BOX(Ri)

AREA(Pi)AREA(BOX(Ri))

RET(Pi) =

Results

• Original image• Rectangular shapes highlighted• Classified regions• Resultant polygons

• Hardware– AMD Athlontm 3000+– 512MB RAM– Linux Mandriva 2006

1st – Original

Over-Segmentation

1998 polygons

Classification

trees

roofs

buildings

streets

others

Resultant Re-Segmentation

634 polygons

317 seconds

2nd – Original

Over-Segmentation

2028 polygons

Classification

trees

roofs

buildings

streets

others

Resultant Re-Segmentation

695 polygons

243 seconds

3rd – Original

Over-Segmentation

2264 polygons

Classification

trees

roofs

buildings

streets

others

Resultant Re-Segmentation

750 polygons

196 seconds

Drawbacks

Conclusions

• New approach for image re-segmentation

• Algorithm developed using the Free C++ Library TerraLib (http://www.terralib.org/)

IMAGE RE-SEGMENTATIONA new approach applied to Urban Imagery

END

Recommended