Image Processing Reference
In-Depth Information
Fig. 10.2 Two dimensions of 'homogeneity' as implemented in eCognition
defined criteria of homogeneity through a combination of parameters: scale (i.e.
average size of objects), color (i.e. mean spectral value) and shape (i.e. geomet-
ric form of the objects) (Fig. 10.2 ).
While the scale parameter controls the average size of the generated objects, the
color and shape parameters reflect the respective criteria of homogeneity being
applied. The software allows absolute and relative weighting of these parameters,
and in addition it splits the shape parameter into outline smoothness and shape
compactness. It helps to extract different object shapes, each representative of dif-
ferent categories in the respective level. Based on these per-level parameters, a
hierarchical network of image objects can be built, which allows for the simultane-
ous representation of image information at different resolutions ('scales').
Object-Based Classification
Object Features
After performing multi-scale segmentation the original image is represented by image
objects on different levels of detail, organized into a hierar-
chical network. The feature space for the description of image
objects is characterized by a vast set of available measures.
Image objects are described by spectral, textural and contextual
measures that are stored in an object feature database. Selected
features are then used to subsequently classify objects.
image objects
are described by
spectral, textural
and contextual
Objects generated through segmentation can be classified using two different clas-
sification methods: (1) classification based on samples and (2) classification based
on the integration of prior external knowledge stored in rule bases.
Search WWH ::

Custom Search