Page 167 - Digital Analysis of Remotely Sensed Imagery

Image Pr ocessing Systems 133

4.6 eCognition
eCognition was developed by Definiens in order to overcome the
weaknesses and limitations of traditional spectral image classifica-
tion methods for hyperspatial resolution satellite imagery inherent in
current image analysis systems. It represents an attempt to remedy
the deficiency of the conventional per-pixel image classifiers that
treat images as composed of individual pixels instead of meaningful
objects. Based on the assumption that objects provide important
semantic clues critical to properly labeling image pixels, eCognition
adopts an object-oriented and multiscale approach toward image
analysis. This attempt provides a powerful means for analyzing
images in an effort to achieve more reliable classifications.

4.6.1 General Overview
The flagship package of Definiens is eCognition Professional, an
object-oriented image classifier that takes contextual information
into the decision making behind classification (Definiens, 2008). In
addition to tone that is the sole clue in spectral classification, shape,
texture, area, and contexture, are all used in eCognition classifica-
tion. As a consequence of this innovation, complex image data are
classified more intelligently, more accurately, and more efficiently
than with traditional methods in a number of steps, ranging from
multisource data fusion, multiresolution image segmentation, and
fuzzy classification. Data from a wide variety of sensors and plat-
forms (e.g., Landsat and Radarsat) of different spatial and spectral
resolutions (e.g., IKONOS and SPOT) may be merged. Image analy-
sis starts with segmentation of remote sensing images into homoge-
neous objects. General knowledge of object features is applied to
improving the accuracy of identification. Image classification is
implemented through sample objects (training areas) or the knowl-
edge base. Sample-based fuzzy classification is a very simple, rapid
supervised classification. Knowledge-based fuzzy classification
relies on knowledge about the relevant image content (e.g., contex-
ture) stored in a knowledge base.
Apart from eCognition Professional, Definiens offers three
other eCongnition packages: Elements, Enterprise, and Forester.
eCognition Elements offers a subset of the image processing tools
in eCognition Professional for multisource data fusion, multireso-
lution image segmentation, and supervised fuzzy classification. It
is hence less complex and easier to use than eCognition Profes-
sional. eCognition Enterprise is an expanded version of eCogni-
tion Professional with extra enhanced functionality for increasing
the efficiency of image classification through batch processing.
This modular client-server system is developed for server-based
image classification to centralize enterprise image classification