Image Processing Reference
In-Depth Information
a certain width. In the object-oriented approach segments represent objects of interest
in different area sizes. Therefore the segmentation parameter with various manipu-
lations is crucial for successful texture analysis and the following classification.
The resulting mask of the city footprint is used to differentiate between sealed and
unsealed areas within the city border.
Chapter Summary
The main idea of object-based image analysis is to work on homogenous image
objects rather than on single pixels. Using spectral and spatial information the
pixels are merged into homogenous groups (segments, image objects). After
segmentation the features of generated objects are used for image interpretation
and subsequent classification. Thus the user's knowledge can be integrated
into image processing and the potential set of target classes can be extended.
There are several segmentation techniques available, with this chapter
focusing on the algorithms implemented in eCognition. These allow image
segmentation into an arbitrary set of hierarchical scale levels, grouping pixels
according to a scale parameter and two aspects of homogeneity: color and
shape/form. Object-based classification utilizes external knowledge by means
of a rule base. Intuitive knowledge can be formalized and made operational
through fuzzy membership functions. Two case studies demonstrate practical
applications of object-based image analysis in an urban context.
LEARNING ACTIVITIES
Internet Resources for Segmentation Approaches and Software Packages
•
Definiens Professional/eCognition: Software based on a region growing
algorithm.
http://www.ecognition.com
Feature Analyst (Visual Learning Systems): Extension for ESRI's ArcView and
•
ArcGIS and ERDAS' IMAGINE to extract object-specific geographic features.
http://www.featureanalyst.com/
PARBAT:
•
http://parbat.lucieer.net
