Segmentation IDRISI Focus Paper
3
Unlike traditional pixel-based classication methods, segment-based classication is an approach that classies a remotely-sensed image based on image segments. Segmentation is the process o dening homogeneou s pixels into these spec trally similar image segments. Te goal o the se gmentation process is to change the characteristics o the image into more meaningul ones, t hus acilitating interpretation and classication . Because these image segments better represent objects in the landscape than do the original pixels, each step o the classication process, rom dening training sites to classiying rom these segments, is simplied. I t is also possible to achieve better accuracy. Te common salt-and-pepper eect that results rom a pixel-based classication is reduced and a more cartographic-grade map is the result. Segment-based classication is highly suited or applications that utilize medium to high resolution satellite imagery and is a useul addition or those mapping land cover and monitoring land change. Other applications such as biodiversity and habitat mapping, where a lone pixel in the classication result may not t in its context, can also take advantage o t his classication appr oach. Tis paper explores how this unctionality is incorporated within IDRISI and also outlines the workfow. Utilizing a powerul set o existing classication tools, rom the traditional maximum likelihood to the cutting edge machine learning tools such as the multi-layer perceptro n and classication tree analysis, IDRISI’ s methodology combines pixel-based and segment-based approaches . Tree modules have been developed to acilitate the creation o a segment-based classied map. SEGMENAION creates an image o segments. SEGRAIN interactively develops training sites and signatures. SEGCLASS classies the image utilizing a majority rule algorithm. image segmentation Te SEGMENAION module generates an image o segments where pixels identied within a segment share a homogeneous spectral similarit y. Across space and over all input bands, a moving window assesses this similarity and segments are dened according to a user- specied similarity threshold. Te smaller the threshold, the more homogeneou s the segments. A larger threshold will cause a more heterogeneous and generalized segmentation result. IDRISI utilizes a watershed delineation approach or the initial p artition o the input images. Ten, similar segments are merged to orm larger segments, based on a user-specied similarity thr eshold. Te segmentation process consists o three procedures: 1. Derive surace image First, or each input image, a correspondin g variance image is cr eated using a user -dened lter . As an example, pixels that are more homogeneous will be assigned lower variance values, while pixels at the boundaries o homogeneou s regions will be assigned higher values. I there is more than one input image, the nal surace image will be a weighted average o all variance images r om all image layers . Te weight o each image is specied by the user . 2. Delineate watersheds Te pixels’ values within the variance image are then treated like elevation values within a digital elevation model. Pixels are grouped into one watershed i they are within one catchment. Each watershed is given a unique integral value as its ID. 3. Merge watersheds Te watersheds or image segments are t hen merged to orm new segments. Te process is guided by the ollowing logic: Segmentation and Segment-Based Classifcation IDRISI Focus Paper Copyright 2009 Clark Labs The SEGMENTATION module creates an image o segments that have spectral similarity across many input bands. This example shows two levels o segmentation rom hal-meter 4-band aerial photography in the blue, green, red and near inrared wavelengths or Woburn, Massachusetts in 2005. The image on the let uses a larger similarity threshold than the one on the right, resulting in more generalized, less homogeneous segments. Using this threshold, the image allows or segments that wholly contain building eatures.
Transcript of Segmentation IDRISI Focus Paper
7/31/2019 Segmentation IDRISI Focus Paper
http://slidepdf.com/reader/full/segmentation-idrisi-focus-paper 1/2
Unlike traditional pixel-based classication methods, segment-based
classication is an approach that classies a remotely-sensed image
based on image segments. Segmentation is the process o dening
homogeneous pixels into these spectrally similar image segments.
Te goal o the segmentation process is to change the characteristics o
the image into more meaningul ones, thus acilitating interpretation and
classication. Because these image segments better represent objects in
the landscape than do the original pixels, each step o the classication
process, rom dening training sites to classiying rom these segments,
is simplied. It is also possible to achieve better accuracy. Te common
salt-and-pepper eect that results rom a pixel-based classication is
reduced and a more cartographic-grade map is the result.
Segment-based classication is highly suited or applications that
utilize medium to high resolution satellite imagery and is a useul
addition or those mapping land cover and monitoring land change.
Other applications such as biodiversity and habitat mapping, where a
lone pixel in the classication result may not t in its context, can also
take advantage o this classication approach. Tis paper explores how
this unctionality is incorporated within IDRISI and also outlines the
workfow.
Utilizing a powerul set o existing classication tools, rom the
traditional maximum likelihood to the cutting edge machine learning
tools such as the multi-layer perceptron and classication tree analysis,
IDRISI’s methodology combines pixel-based and segment-based
approaches. Tree modules have been developed to acilitate the creation
o a segment-based classied map. SEGMENAION creates an image o
segments. SEGRAIN interactively develops training sites and signatures
SEGCLASS classies the image utilizing a majority rule algorithm.
image segmentation
Te SEGMENAION module generates an image o segments where
pixels identied within a segment share a homogeneous spectral
similarity. Across space and over all input bands, a moving window
assesses this similarity and segments are dened according to a user-
specied similarity threshold. Te smaller the threshold, the more
homogeneous the segments. A larger threshold will cause a more
heterogeneous and generalized segmentation result.
IDRISI utilizes a watershed delineation approach or the initial partition
o the input images. Ten, similar segments are merged to orm
larger segments, based on a user-specied similarity threshold. Te
segmentation process consists o three procedures:
1. Derive surace image
First, or each input image, a corresponding variance image is created using a user-dened lter.
As an example, pixels that are more homogeneous
will be assigned lower variance values, while pixels
at the boundaries o homogeneous regions will be
assigned higher values.
I there is more than one input image, the nal
surace image will be a weighted average o all
variance images rom all image layers. Te weight o
each image is specied by the user.
2. Delineate watersheds
Te pixels’ values within the variance image arethen treated like elevation values within a digital
elevation model. Pixels are grouped into one
watershed i they are within one catchment. Each
watershed is given a unique integral value as its ID.
3. Merge watersheds
Te watersheds or image segments are then merged
to orm new segments. Te process is guided by the
ollowing logic:
Segmentation and Segment-Based Classifcation
IDRISI Focus PaperCopyright 2009 Clark Labs
The SEGMENTATION module creates an image o segments that have spectral similarity across many input bands. This
example shows two levels o segmentation rom hal-meter 4-band aerial photography in the blue, green, red and near
inrared wavelengths or Woburn, Massachusetts in 2005. The image on the let uses a larger similarity threshold than
the one on the right, resulting in more generalized, less homogeneous segments. Using this threshold, the image allows
or segments that wholly contain building eatures.
7/31/2019 Segmentation IDRISI Focus Paper
http://slidepdf.com/reader/full/segmentation-idrisi-focus-paper 2/2
classification
Te last step in segment-based classication is to classiy the image
segments. Tis is done with the assistance o a reerence image. A
reerence image is an already classied image. Tis image can be createdusing the segment-based signatures or some other approach. Te
important point here is that this reerence image is used to assign the
majority class within each segment.
A majority rule algorithm is applied within each segment to determine
its class assignment. Since the base unit in a segment-based classication
is an image segment, the accuracy may improve upon the pixel-based
classication and produce a cartographic-grade classication result as
well.
Te segment-based classication approach within IDRISI is
straightorward and bypasses the complex rule-base construction
process ound in other tools. It also provides the user with a variety o
classication choices or the creation o the reerence image. Te optionalso exists or combining the many pixel-based classiers with this
approach to create a hybrid classication procedure not ound elsewhere.
Clark Labs
Clark University
950 Main Street
Worcester, MA 01610-1477 USA
tel:+1.508.793.7526
ax: +1.508.793.8842
email: [email protected]
www.clarklabs.org
-- Te segments must be adjacent and mutually most
similar.
-- Te dierence between the mean and standard deviation
values o the two segments within a pair must be less than
the user-specied threshold. A fexible tuning mechanism
is included to allow one measurement to have more weight
over the other i appropriate. Te threshold, or similarity tolerance, controls the generalization level. Te larger the
value, the ewer number o segments in the output.
training sites and signature development
Te next step in the classication process is to derive
training sites and signature classes rom the image
segments. In addition to the current signature development
tools, the SEGRAIN module creates training sites and
signature classes based on the image segments. With an
intuitive graphical interace, you can interactively select
segments o interest as sample sites or particular classes.
Once a segment is selected, all pixels in that segment areused or signature development. Tese images can then be
used as input into one o IDRISI’s many classiers.
The module SEGCLASS classifes the imagery using a majority rule algorithm to assign each segment to the
majority class rom the reerence image. SEGCLASS can improve the accuracy o a pixel-based classifcation and
produce a smoother map-like classifcation result while preserving the boundaries between the segments.
The module SEGTRAIN assigns these segments to specifc land cover types or the
development o training site data. The user interactively selects segments and assigns class
IDs and class names.
