Image Processing Reference
In-Depth Information
10.4
Case Studies
Remote sensing in urban areas covers a huge spectrum of applications and scientific
research, where object-based analysis can be very supportive. In this chapter two
case studies are briefly presented, one based on the detection of informal settle-
ments, the other on textural analysis for mapping urban structure.
10.4.1
Detecting Urban Features from IKONOS Data Using
Methods of Object Oriented Image Analysis (Hofmann
2001
)
Most urban image classifications target formal settlements like big cities, which are
already studied in detail and mapped extensively. The detection of informal settle-
ments is a challenging task due to their microstructure and irregularity in object
shapes. Based on spectral and spatial resolution two different levels of detail can be
examined: (1) the detection of single shacks or (2) the location of entire informal
settlements and their boundaries.
The study area, a part of Cape Town (South Africa), is characterized by several
different forms of settlement areas. The first step was an image enhancement by
applying a principal-component pan-sharpening method. Next, multiple hierarchi-
cal image object levels were created in eCognition. The smallest level of image
segments reveals single houses, while the top-most level represents entire settle-
ment areas or parts thereof. After conducting a nearest-neighbor classification the
result was improved by using inheritance mechanisms and form criteria. Textural
information described by reflectance and shape of lower-level objects helps to
identify and classify different types of settlement areas.
Hofmann argues that the object-based approach is well suited to detecting the
complex structures of informal settlements. Especially when working with enhanced
high resolution satellite imagery, image segmentation identifies 'real-world' objects
that show a typical texture according to the different types of settlements.
10.4.2
Analysis of Urban Structure and Development. Applying
Procedures for Automatic Mapping of Large-Area Data
(De Kok et al.
2003
)
The analysis of urban structure starts with the central role of texture analysis
for city footprint extraction. In general, texture can be expressed by calculating
Grey Level Co-occurrence Matrices (GLCM). A bottleneck in textural applica-
tions is the relationship between the object of interest and a fixed filter area size
used for the moving window processing. Contrary segment-size is not restricted to
