Image Processing Reference
In-Depth Information
Fig. 10.3
Principle workflow of object-based image analysis
10.3.4
Accuracy Assessment
Quantitative site-specific accuracy assessment (Congalton and Green
1998
) is an
important means to evaluate the quality of classification results. Error matrices and
corresponding assessment values (error of commission,
error of omission, kappa statistic) are frequently used for
evaluating pixel-based classifications. The thematic assess-
ment can be performed by generating random points within
objects and checking the classification labels against a
ground truth layer. Alternatively a specific set of randomly
selected objects can be chosen as test areas (similar to
training areas) and then inspected as reference information.
Geometrical fitting is however by far harder to evaluate.
While the classified image objects can be visually checked against manual delinea-
tion of interpreted imagery (Koch et al.
2003
), the geometrical congruence is
assessed by comparing existing vector data with overlay techniques.
a challenging
issue in assessing
the accuracy of
object-based
classification is
the geometrical
fitting of objects
