eCognition Evaluation

1
eCognition Evaluation

• Object oriented image analysis

2
Presentation Overview

• eCognition Overview (Main Concepts) (10-15min.)
• Demo of test cases (Landsat, SAR, IKONOS data)
  (30-45 min.)
• Evaluation results
• Discussion
• (if time remains User Hands-On)

3
eCognition - object oriented image analysis

• Definiens Imaging, Munich (http//www.definiens-im
  aging.com/)
• Pure Classification tool (not general img.
  Processing Analysing tool as ERDAS)
• 2 month license for testing

4
eCognition Main Concept

• 1. Segmentation (-gt Objects)
• 2. Definition of Classes (Categories)
• e.g. Class Water All objects with in a certain
  spectral mean value range
• 3. Classification of Objects Object classified
  as Class Water ? Object fulfils criterion(s) of
  class definition

5
eCognition - Segmentation

• Merging Algorithm Generation of homogeneous
  image regions (objects)
• 4 Parameters influence the segmentation Scale
  Parameter, Color/Shape Criterion,
  Compactness/Smoothness Criterion, weights of
  image layers
• Parameters are not absolute Indicators (to have
  to try diff. combinations to find appropriate
  segmentation)

6
eCogniton - Pixels vs Objects

• Object attributes Pixel
  attributes
• - Color Statistics - Color
• - Form/Shape
• - Area/Size
• - Texture
• - Context

7
eCognition - Workflow

• Data Input
• Multiresolutuion Segmentation (hierarchical net
  of image objects)
• Creating Class Hierarchy
• Classification

8
eCognition - Classification

• Sample based Classification
• (Test samples, Definition of n- dimensional
  feature space, Nearest Neighbour Classifier)
• Rule based Classification
• (Formulation of 1 dimensional membership
  functions, Neighbourhood relations)

9
eCognition supported Formats

• Raster, Vector, .tif, .gif , .bmp, .img,
  .shp, Arc/Info
• Images need not to be geocoded but different
  Resolutions, different geographic areas, if
  images are geocoded

10
eCognition System Requirements

• eCognition Version 3.0 Operating Systems
  Windows NT 4.0, Windows 2000, Windows XP (not on
  Windows 95/98/ME)
• PC with at minimum 512 MB RAM and Pentium III

11
Test case IKONOS data

• Panchromatic Ikonos Image of area around ESRIN
• Classification of olive groves

12
Test case SAR data

• ERS SAR time series (01/99 01/00)
• CAT TIEN national park in Vietnam
• Estimate the wetland extension over the year
• Showing Changes between 09/09/99 and 21/12/99
• Difference to ENVI thresh hold classification
  (2-3)

13
Test case Landsat data

• Landsat 7 TM (16/08/00), Area around Rome and
  Castelli Romani
• Extraction of river and lakes

14
eCognition Conclusion 1/2

• segmentation process (on different scales),
  relations to neighbour-, sub-, superobjects
• data fusion (different sensor, resolution,
  raster, vector data)
• re-usable semantic models (classes)
• feature analysis tools (spectral, shape,
  neighbourhood relation features)
• support of standard image formats ((geo) tiff,
  jpg, erdas imagine, GIS formats, shape files,
  ARC/INFO)

15
eCognition Conclusion 2/2

• not easy to use (many features, Finding the Scale
  Parameter (Segmentation), new approach, high
  initial learning curve)
• Automation (only macros, no batch mode)
• Segmentation time/memory critical
• Pure supervised approach (class modelling
  critical)