E Cognition User Summit2009 S Lang Zgis Object Validity

Object validity for high-standard
information products
Between theoretical foundations and
operational use
Stefan LANG1,2
1Centrefor Geoinformatics,
University of Salzburg
2Department of Geoinformation Processing
for Landscape and Environmental Planning
Technical University Berlin

eCognition User Summit
3 Nov 2009, Deutsches Museum, Munich

Centre for Geoinformatics (Z_GIS)
 Experts for the spatial view
 Analyse, understand and visualise spatial phenomena, their causes
and mutual relationships as well as their temporal dynamics
 Impart knowledge by teaching and training activities, foster capacity
building
 Establish tight links between GI science, industry and administration
providing platforms for communication and exchange

Knowledge Cycle

2009 Definiens User Summit | S. Lang | „Object validity …“ | 3 Nov 2009 | www.zgis.at/research

2

Academic Centre of Excellence
(ACE) Z_GIS
 ACE Z_GIS
 1stACE for object-based image analysis (OBIA).
 Since the year 2000, Z_GIS is widely recognized as a promoter of the OBIA approach with a
broad range of activities in research, development and training.
 organised and hosted the first International Conference on Object-based Image Analysis in
2006 (OBIA 2006),
 supported last year's GEOBIA 2008 hosted at University of Calgary and upcoming GEOBIAs
 edited the first comprehensive book on Object-based Image Analysis, published by Springer in
2008

 Specifically, the ACE Z_GIS focuses on
 Fundamental and applied research in methods and technologies aiming for a useful
exploitation of remote sensing data in combination with geoinformation
 Applied research in environmental and security-related applications
 Scientific guidance to the dissemination of innovative approaches through education,
awareness raising and new learning technologies
 Integration of Definiens software in higher education programmes in order to broaden the
base of students and future application specialists
 Further development in specifies fields such as forestry, spatial ecology, security research,
monitoring systems in general, etc.
 Supporting improvements in the interface and conceptualization of the software


3

Challenges

 OBIA opens the door
 for entering new geographical realities within multiple
scale domains and complex class descriptions
 Enabling elaborated multi-scale object representations
and class modelling for addressing
 composite bona fide objects or
 even concept-related fiat objects
 By this, OBIA poses challenges
 for evaluating object validity … in both scientific and
operational context


Object validity

 The multi-scale solution …
 (1) scaled representations must correspond to an
underlying theoretical or ontological framework;
 (2) user requirements of suitability and reliability must
be met;
 (3) qualification for existing workflows and geospatial
infrastructures must be assured.


Validation and use
Usability
(accuracy,
Object validity
stability,
transferability,
Class modelling /
level of cognition network
automation)
Spatial analysis

Epistemological
significance
Object fate
Decomposability Pattern recognition

Hierarchy theory Machine-based image analysis

(multi-scale) (automated)
Production system,
Segmentation classification rule bases
Image object
Scaled representation
(generalization) Image
Pixel aggregate understanding
Image region
‘Gestalt’ problems
Visual perception / Classification
human expert / schemes, experiences, expert
user instance knowledge, existing / external
geometryS

6

3rd match
Image understanding and OBIA
Semantic system  conceptual reality

Spatial
distribution of
Scene description Transparency
Transferability
categorized as conceptual reality Objectivity
objects
Utilization and transformation

2nd match
Spectral,
of knowledge

Object
hypotheses
Class Modelling structural,
semantic
categorizing image objects
characteristics and relationships Object and class
Class system modelling

1st match Segmentation
problem
Domain of Knowing what we are looking for …
interest
target objects, scales and classes
scale (dependent on the domain of interest)
Target objects multi-scale representation Complex scene
classes content


7

Regionalised hierarchies

From Tiede et al, 2004
From Lang, 2002


8


Different patterns of
increasing object size
with incremental multi-
res. segmentation

From Lang, 2002

9

ESP tool taking into account
local variance among
objects and rate of change
between levels

From Dragut et al, in press

10

OFA and OBAA

 Spatial relationships among corresponding geographic2008
Fig: Albrecht

objects
 From different (scaled) representations (e.g. manual
delineation vs. segmentation) or different time slices

Scale-specific Scale-adaptive
representation representation

Comparison visual interpretation vs. automated, machine-based delineation of
habitats (left: FCIR air-photo, 25 cm GSD, courtesy of FH Weihenstephan;
right: QuickBird imagery pansharpened 0.6m GSD)

OFA and OBAA

 Object fate analysis OFA
 Object-based accuracy assessment (OBAA)
T0 two different time slices
T0 and T1 with the same
T1 representation, e.g. R0
Corresponding
Layer

R0 two different
OFA
representations R0 and R1
R1 at the same time slice,
Reference e.g. T0
layer

T0 R1
Reality: combination, i.e.
Lang, 2008; Schöpfer et al. 2008; Albrecht 2008 different representations
T 1 R1
from different time slices


OFA and OBAA
Increasing level of interaction

Fig: Albrecht in Schöpfer et al 2008
Object interior Object interior Object interior,
and boundary boundary and exterior

Good II

Good I Expanding
C2
similar to “equal“

complementing
categories

C1
similar to “disjoint“
Invading
Not interfering I n exp
R e _ ge
n good n exp


Towards high-level information products
Geospatial Information
products for
explaining, visualizing and
monitoring
complex spatial Disaggregated
population data
phenomena within
contamination
zones of fictitious
plume
(Lang / Tiede)

OBIA allows combining of
• data integration automated image
analysis and information extraction
• regionalization and multi-scale
representation
• spatial analysis, change detection and
modeling

14

Information on land-use transformation
for regional development policy
Satellite data
-SPOT 5 (5m-color) mosaic
 Demand profile and policy scope Auxiliary in situ data
- DEM
 Verband Region Stuttgart - ALK digital cadastral data,
- Biotope mapping (hint layer)
 Biotope complexes as basic units for - ATKIS digital topographic data
regional planning purposes
 BIMS (Biotope information and
management system)

VII Arable land, poor in 4 ha
accompanying habitat
structures
VIII Arable land, rich in 2 ha
accompanying habitat
structures

XI Mixed arable land and 2 ha
grassland area
Stuttgart Region (approx. 3600 km²)
XII Agriculturally improved 2 ha
grassland
50 km


• Geometric conditioning (individual
treatment of cadastre boundaries)
 (1) boundaries retained: parcel
corresponds to one, single homogenous
image object, no change or update
 (2) boundaries removed: parcels merged
because of internal homogeneity,
change of geometry
 (3) boundaries introduced: single parcel is
spectrally heterogeneous and split
according to spectral behaviour, change of
geometry (mainly forest)
Lang et al, 2007, Tiede et al., 2007



 Thematic conditioning (functionally homogenous
units, minimum size)
 31,698 biotope complexes were delineated for the whole
Stuttgart Region
 Average size: 11.5 ha

Lang et al, 2007, Tiede et al., 2007



Automatically composed

Elementary units

… How to validate?
Manually composed

18

Information on IDP camp growth
for post-conflict disaster management

Automatically extracted dwelling units 
Calculation of density zones 
development over time


19

Other multidimensional spatial
phenomena …

e.g. Vulnerability to flood hazard (Kienberger et al., 2009)


20

Sensitivity: high-level indicator for SEA

Source: Kienberger et al, 2009


Example vulnerability units

 Decomposition of vulnerability into domains and indicators
 Regionalisation are applied to derive discrete vulnerability units
 Weighting according to relative importance (expert knowledge)

Algorithm after
Baatz & Schäpe
(2000)
Kienberger et al., 2009


OBIA – Meeting the challenges

 After Kuhn (1962), data scarcity is a characteristic of paradigm
shift  no: data abundance, affluences, but lack of solutions
 Algorithms, methods and technology (rule-based vs. adaptive
learning, genetic algorithms, Markov chains ...) to be fused (?)
 From image understanding to problem understanding (user-
oriented)
 Make products ready for further analysis, ready to be
integrated in daily workflows
 Tackle challenges of the world (monitoring, ...).
 Are we using the proper tools for solving specific problems
(segmentation, generalisation, cartography)


OBIA – Strengthening the strengths

 Multi-multi-dimensionality (multi-disciplinary, multi user
profiles, multi application)  Network, “trans”
 Tight linkages between industry, research and academia.
 How do we turn data to information? Listen to the customers
/ users.
 Obey efforts in data models, standardization
 Validation / benchmarking: Not only data become more
complex, but also the automation process
 Common vocabulary, ontology, reflecting on ideas
 Sustainable use … teach and multiply it! Make students
become acquainted early enough


Thank you very much!

stefan.lang@sbg.ac.at

Contact: stefan.lang@sbg.ac.at

Activities were been carried out drawing from various funding sources, in
the framework of the Salzburg Academic Centre of Excellence SACE.

E Cognition User Summit2009 S Lang Zgis Object Validity

More Related Content

What's hot (20)

Viewers also liked (13)

Similar to E Cognition User Summit2009 S Lang Zgis Object Validity (20)

More from Trimble Geospatial Munich (11)

Recently uploaded (20)

E Cognition User Summit2009 S Lang Zgis Object Validity