An object-oriented pattern recognition approach for urban classification - Urban Remote Sensing: Monitoring, Synthesis and Modeling in the Urban Environment - page 125

TABLE 9.1 Available features belong to Generic sub-category under Shape category.

Feature Type Category Sub-category Features

1. Object features 1.1 Customized

1.2 Layer Values
1.3 Shape 1.3.1 Generic 1.3.1.1 Area

1.3.1.2 Asymmetry
1.3.1.3 Border Index
1.3.1.4 Border Length
1.3.1.5 Compactness
1.3.1.6 Density
1.3.1.7 Elliptic Fit
1.3.1.8 Elliptic Fit (legacy feature)
1.3.1.9 Length
1.3.2.0 Length/Width
1.3.2.1 Main Direction
1.3.2.2 Radius of Largest Enclosed Ellipse
1.3.2.3 Radius of Largest Enclosed Ellipse(legacy feature)
1.3.2.4 Radius of Smallest Enclosed Ellipse
1.3.2.5 Radius of Smallest Enclosed Ellipse (legacy feature)
1.3.2.6 Rectangular Fit
1.3.2.7 Rectangular Fit (legacy feature)
1.3.2.8 Roundness
1.3.2.9 Roundness (legacy feature)
1.3.3.0 Shape Index
1.3.3.1 Width

1.3.2 Position
1.3.3 To super object
1.3.4 Based on Polygons
1.3.5 Based on Skeletons

1.4 Texture
1.5 Variables
1.6 Hierarchy

(1) identify certain image objects as samples, and (2) classify
image objects due to their nearest sample neighbors in feature
space (Definiens, 2008). The nearest neighbor classifier is also
based on a non-parametric rule and is therefore independent of
normally distributed input data. The nearest neighbor approach
enables transportability of a classification system to other areas,
requiring only the selection or modification of new objects
(training samples) until a satisfactory result is obtained (Ivits
and Koch, 2002). Application of the nearest neighbor method
is also advantageous when using spectrally similar classes that
are not well separated using a few features or just one feature
(Definiens, 2008).
There are two options available with the nearest neighbor
function in Definiens software, namely (1) Standard Nearest
Neighbor, and (2) Nearest Neighbor. The Standard Nearest

Neighbor option automatically select mean values of objects for
all the original bands in the selected image, whereas the second
option requires users to specify selected features (e.g., shape,
texture, hierarchy).

9.3 Data and study area

We used two subsets of a Quickbird image data over the city of
Phoenix acquired on 22 August 2007 to demonstrate how deci-
sion rules and nearest neighbor classifiers can be used to identify
land cover classes using an object-oriented approach. Figure 9.2
presents the first subset used for different decision rules to extract
swimming pools and Figure 9.3 shows the second subset used