UNSUPERVISED CLASSIFICATION
Mirza Muhammad Waqar
Contact:
[Link]@[Link]
+92-21-34650765-79 EXT:2257
RG610
Course: Introduction to RS & DIP
Contents
2
Satellite Image Classification
Spectral vs Spatial Pattern Recognition
Feature Space
Image Spectral Signature
Limitations in Classification
Unsupervised Classification
Clustering Algorithms
K-means
Approach
ISODATA Model
Texture or Roughness Model
Post Classification
Satellite Image Classification
Automatic categorization of all pixel into land
cover classes.
To explore different land covers in a satellite
image.
To assign unique no or symbol to particular
land cover.
Qualitative and Quantitative analysis of
satellite image.
Spectral vs Spatial
Spectral pattern recognition
Family of classification procedures that utilize
pixel by pixel spectral information as the
basis for automated land cover classification.
Spatial pattern recognition
Categorization of image pixels on the basis of
their spatial relationship with pixels
surrounding them.
Feature Space
Feature Space
Image Spectral Signatures
Satellite Image
Classification
Three types of classifications
Un-Supervised Classification
Supervised Classification
Hybrid Classification
Limitations in Classification
Limitations face in satellite image
classification
Pixel Size
All Statistical parameters are developed for
normalized distribution.
Un-Supervised Classification
Use to cluster pixels based on statistics only.
No user defined training classes required.
Machine based classification.
Post classification is of more importance to
make results meaningful.
Incorporate all the natural groups in satellite
image (spectral classes).
Un-supervised Classification have two phases.
Clustering
Post Classification
Clustering Algorithms
Numerous clustering algorithms
K-means Approach
ISODATA Model
Texture or Roughness Model
K-means Approach
Accept number of clusters to be located in
the data.
Arbitrarily locate that number of cluster
centers in multi-dimensional
measurement space.
Each pixel is assigned to the cluster whose
mean vector is closest.
B
a
n
d
B
Band A
B
a
n
d
New
computed
Means
Previous
Means
Band A
B
a
n
d
New
computed
Means
Previous
Means
Band A
K-Mean Approach
New means are computed.
Revised clustering on the base of new
computed means.
This process continue until there is no
significant change in clusters mean.
ISODATA Model
Iterative Self-Organizing Data Analysis.
Follow K-mean principle for clustering.
Accept number of clutters, number of
iterations & convergence tolerance from the
user and form clusters.
Permits number of clusters to change from
one iteration to the next, by merging,
splitting and deleting clusters based on
spatial statistics and user defined
conditions.
B
a
n
d
New
computed
Means
Previous
Means
Band A
Texture or Roughness Model
It incorporate a sensitivity to image
texture or roughness as a basis for
establishing clusters centers.
Texture is computed through multidimensional variance observed in moving
window (e.g. a 3x3 window).
Analyst sets a variance threshold below
which window is consider smooth and
above which it is considered rough.
Texture or Roughness Model
The mean of the first smooth window
encountered in the image becomes the
first cluster center.
The mean of the second smooth window
encountered in the image becomes the
second cluster center and so forth.
This process continue until the user
defined no of clusters reached.
Post Classification
In post classification phase, analyst compare
spectral classes with some reference data to
determine the identify of the spectral classes.
Spectral reflectance curves can be used to
identify the spectral classes.
Defining the level of classification
Merging different classes to reach final
outcome.
Accuracy assessment through field truthing.
Spectral Classes
Identity of Spectral
Class
Corresponding
Desired Information
Category
Possible outcome 1
Water
Water
Coniferous trees
Coniferous trees
Broad Leave trees
Broad Leave trees
Bare Soil
Bare Soil
Rocks
Rocks
Built-up Area
Built-up Area
Water
Water
Coniferous trees
Forest
Broad Leave trees
Bare Soil
Rocks
Roads
Urban Area
Possible outcome 2
Open Land
Built-up Area
Questions & Discussion
