Todays Schedule

1
Todays Schedule

• 830-900 am Continental breakfast
• 900-910 am Introductions and welcome
• 910-1040 am Workshop 1 Software Options in
  OBIA
• 1040-1100 am Break
• 1100-1230 pm Presentation Session 1 Land Cover
  / Arid Environments
• 1230-130 pm Lunch
• 130-300 pm Workshop 2 Advanced Segmentation
  with Definiens eCognition
• 300-330 pm Break
• 330-500 pm Presentation Session 2 Forest
  Landscapes
• 500-530 pm Summary of the day
• 600-900 pm Keynote Dinner

2
Workshop 1Software Options for OBIA

• Instructor Marek Jakubowski
• (marekj_at_nature.berkeley.edu)
• OBIA Symposium
• June 7-8, 2007
• Geospatial Imaging Informatics Facility
• http//giif.cnr.berkeley.edu

3
Outline

• Goal To explore available segmentation software.
• Segmentation software
• ImageSeg
• Image Segmentation
• Multiscale Image Segmentation
• CAESAR InfoPACK
• Other
• Hands-on Workshop
• SPRING
• Feature Analyst
• Definiens Professional (eCognition)

4
ImageSeg

• Implementation plug-in for ENVI
• Algorithm Texture-based image segmentation
• Cost Free
• Current version 0.1
• Updates no longer updated
• Severe limitations
• Image size (LW2n)
• Texture only
• Results
• All unique segments
• Poor with RS imagery
• Available at http//www.ittvis.com/codebank/searc
  h.asp?FID361

5
ImageSeg
6
ERDAS Imagine Image Segmentation

• Implementation plug-in
• Algorithm texture based
• Cost Free
• Updates no longer updated (8/2004)
• Results over segments RS imagery
• Available at Leicas website (sometimes)

Meinel, G, M Neubert, 2004. A Comparison of
Segmentation Programs for High Resolution Remote
Sensing Data, Commission VI on Proceeding of
20th ISPRS Congress in Istanbul.
7
ERDAS Imagine Image Segmentation
Meinel, G, M Neubert, 2004. A Comparison of
Segmentation Programs for High Resolution Remote
Sensing Data, Commission VI on Proceeding of
20th ISPRS Congress in Istanbul.
8
Multiscale Image Segmentation

• Implementation stand alone command line
• Algorithm Edgeflow-driven
• Cost Free open-source
• Updates October 2005
• Results poor with RS imagery
• Available at http//barissumengen.com/seg/

Meinel, G, M Neubert, 2004. A Comparison of
Segmentation Programs for High Resolution Remote
Sensing Data, Commission VI on Proceeding of
20th ISPRS Congress in Istanbul.
9
InfoPACK

• Implementation stand alone application
• Algorithm intended for SAR data
• Cost Expensive
• Current version 2
• Updates roughly annual
• Results over segments RS imagery
• Available at http//www.infosar.co.uk/misc/produc
  ts.html

Meinel, G, M Neubert, 2004. A Comparison of
Segmentation Programs for High Resolution Remote
Sensing Data, Commission VI on Proceeding of
20th ISPRS Congress in Istanbul.
10
InfoPACK
Meinel, G, M Neubert, 2004. A Comparison of
Segmentation Programs for High Resolution Remote
Sensing Data, Commission VI on Proceeding of
20th ISPRS Congress in Istanbul.
11
Other options

• Qinghua Guo
• EDISON Edge Detection and Image Segmentation
• Peng Gong Astro
• Geoffrey J. Hay SCRM
• multiscale object-based

http//faculty.ucmerced.edu/qguo/index.htmlhttp/
/www.cnr.berkeley.edu/gong/http//homepages.ucal
gary.ca/gjhay/people/Geoff_new/geoff_projects.htm
12
Outline

• Goal To explore available segmentation software.
• Segmentation software
• ImageSeg
• Image Segmentation
• Multiscale Image Segmentation
• CAESAR InfoPACK
• Other
• Hands-on Workshop
• SPRING
• Feature Analyst
• Definiens Professional (eCognition)

13
SPRING

• Implementation stand-alone application
• Algorithm Region growing
• Cost Free
• Current version 4.3.2
• Updates unknown
• Advantages
• Cost
• Disadvantages
• Image support
• Output
• Running time
• Results
• Well suited in urban environments
• Over segments in natural environments
• Available at http//www.dpi.inpe.br/spring

SPRING Integrating remote sensing and GIS by
object-oriented data modelling" Camara G, Souza
RCM, Freitas UM, Garrido J Computers Graphics,
20 (3) 395-403, May-Jun 1996.
14
Similarity 120 Area 500
15
Similarity 80 Area 500
16
Similarity 35 Area 500
17
Feature Analyst

• Implementation plug-in for ERDAS Imagine
  ArcGIS, other
• Algorithm statistical machine-learning feature
  recognition
• Current version 4.1
• Updates frequent
• Cost 1000 annual subscription (educational)
• Available at http//www.featureanalyst.com/

18
Feature Analyst

• Advantages
• Easy to use
• Easy-to-read manual
• Export every step
• Useful tools
• Linear SHP
• 3D extraction
• Automatically classifies

• Disadvantages
• Difficult to use very effectively
• Multiple features problematic
• Many layers ? slow interface
• Mostly supervised
• Extensive user input
• Very dependent on user input
• garbage in garbage out

• Results
• Good for quick information extraction
• Single targets
• Linear networks
• Extensive analysis can be difficult
• Simultaneous multiple features

19
Feature Analyst - Workflow

• Define features (digitize)
• Setup learning
• Start learning algorithm
• Analyze results and adjust features/learning
• Re-run learning
• Export results as desired (raster, polygon, line,
  or point)
• OR
• Unsupervised

20
Feature Analyst - Supervised
21
Feature Analyst - Unsupervised
Pattern width 9
22
Feature Analyst - Unsupervised
Pattern width 5
23
Definiens Professional Earth

• Implementation stand-alone application
• Algorithm fractal-based region growing
• Cost expensive
• Current version 5.0
• Updates frequent
• Available at http//www.definiens.com/

24
Definiens Professional

• Advantages
• Multiscale segmentation
• Transparent classification process
• Object statistics

• Disadvantages
• Steep learning curve
• Often requires trial-and-error
• Difficult to keep track of complicated projects
• Segmentation parameters not intuitive

25
Definiens Professional Example
26
Definiens - current version
Professional
4x Segmentation 3x Classification 4x Advanced
Classification 3x Level Operation 3x Reshaping 2x
Vectorization 3x Sample Operation 2x Thematic
Layer Operation 7x Export 31 Algorithms
27
Definiens - next version
Developer
6x Segmentation 4x Classification 5x Advanced
Classification 5x Variable Operation 3x Level
Operation 6x Reshaping 2x Vectorization 5x Sample
Operation 3x Thematic Layer Operation 9x
Export 7x Training Operations 2x Image Layer
Operations 5x Workspace Automation 62 Algorithms
28
Lab Exercises