Satellite image classification

1
Satellite image classification
SSIP 08, Vienna

• Tamas Blaskovics University of
  Szeged
• Michael Glatz Vienna University
  of Technology
• Korfiatis Panagiotis University
  of Patras,
• José Ramos Porto University

2
Task

• Satellite Image Classification
• Input Landsat images of terrain, plus sample
  images of fields/ sea, forest etc
• Aim segmentation of scene based on texture (and
  color)
• Additional goal intenfication of key features
  such as cave openings etc
• Output labeled scene

3
Satellite image classification
Input Image
MRF Semi- Supervised Segmentation
MRF Unsupervised Segmentation
k-means Unsupervised Segmentation
or
or
Output Image
or
Area classification ( User Interaction)
Area classification ( Automated)
4
Dataset
20 images aquired with the IKONOS
Satellite. (http//www.satimagingcorp.com/satellit
e-sensors/ikonos.html)
5
Method 1/2

• Step 1 Image Segmentation

• The RGB image was converted to Luv color space
• Two unsupervised methods were used
• MRF segmentation ( Kato et al. )
• EM step
• ICM
• K-means
• Parameters
• User defined of regions, ß, temperature.

6
Method 2/2

• Step 2 Class Characterization

• User defined
• User chooses the desired region for
  classification
• The first order statistics (mean, variance,
  skewness, kurtosis, range) are calculated for a
  ROI around the selected image
• Automated
• Skeletonization technique was applied for each
  segmented region
• A sliding ROI (21 x 21) was used to extract first
  order statistics
• K-nearest neighbor classifier was used (NN)
• Segmented area is also calculated

7
Features evaluated
Segmentation Stage

• Intensity value channel U
• Intensity value channel V

Classification Stage

• Mean value
• Standard deviation
• Kurtosis
• Skewness
• Range

8
GUI
9
Result image segmentation 1/2
10
Result image segmentation 2/2
11
Comments

• Visual evaluation seems to present good results
• No serious evaluation was conducted
• Segmentation process is slow
• Dataset is too small to construct robust learning
  process

12
Future developments

• Segmentation process
• Evaluation of more complex techniques
• Classification process
• Bigger training database
• Other texture features
• Different classifiers
• Evaluation
• Use of ground truth and shape differentiation
  metrics

13
THE ENDThank you!