1. Studio 7: Surface Data Capture
Presented by
Rashid Javed (704076)
Supervisor:
Prof. Dr. Ing. Eberhard Gülch

2. AGENDA:
1) Introduction
2) Workflow
3) Data and Software Used
4) Methodology
5) Experiment Results
6) Conclusion

3. Introduction
In the development of digital photogrammetric system, automatic image
matching process play an important role.
 The automatic image matching is used in finding the conjugate points of an
aerial photograph stereo pair automatically.
The first solution of problems to Image matching was given by Hobrough in late
1950.
Main task is to Develop a procedure for automatic point transfer of regular
raster points in an epipolar image pair

4. Introduction
Epipolar Images:
Epipolar images are stereo pairs in which the left and right images are oriented in
such a way that ground feature points have the same y-coordinates on both
images.

5. Introduction
Conjugate Points:
• Common points in the overlap areas of the stereo pair images.
• Also called Tie Points
• The search for correspondence points can be done in 2-D e.g. in a
rectangle oriented along an approximate epipolar line.

6. Introduction
Relationship Between Matching Methods and Matching Entities

7. Introduction
Area Based Matching
• Oldest and simplest matching technique
• Is performed on the basis of grey values
• Compare the grey level of small sub image with its counterpart in the
other image
• Template and Search Window

8. Introduction
Correlation
• Technique used for finding the conjugate points
• Measure the similarity of template and search window computing the
correlation factor
Parallax
• Apparent change in the position of object
• Related to the elevation of the terrain
Applications:
 Image registration, DEM generation, displacement measurements, Target
measurements, Surveillance, Treaty verification

9. AGENDA:
Introduction
Objectives:
• To develop a procedure of automatic point transfer to determine the
conjugate points in an epipolar image pair
• To compare the intensity or gray values of a template
• To asses the quality of the match through using correlation coefficient.
• Develop a parallax field of M*N matrix

10. AGENDA:
Workflow
Select odd size template
in the source image
Matching window in the destination
image with conjugate entities
Computation of co-rrelation
coefficients for different threshold
Repeat above steps for new template location until
all positions to be matched are visited
Analyse for
consistency

11. AGENDA:
Workflow
Location of Template:
• Not to locate in the occluded areas, area with low contrast or repetitive
pattern.
Size of Template
• Increase in size will increase the uniqueness of grey value but also increase
distortion.
• Decrease in the size will amount to less distortion but the uniqueness of
the grey values will be decreased.
Location and size of search window
• Location is important (Epipolar Line)
• size should be large enough to include farthest moving template and small
enough to limit the computational cost of matching

12. Workflow
Acceptance Criteria
•Three threshold values are used
Quality Control
•Accuracy and reliability of conjugate Points.

13. AGENDA:
Workflow
Correlation Coefficient

14. Workflow
Calculation of Parallax
• P = Xleft – Xright
• Where,
• Xleft = position of certain object in left image
• Xright = position of same object in right image

15. Data and Software Used
Images:
• Two epipolar images
• MATLAB
Image Left Image Right

16. Methodology:
Develop algorithm in the MATLAB
Take the input of starting point in left image, template size as well as window
size in the right image.
Select the threshold values for correlation coefficient calculation.
Threshold 1 = 0 ≤ r ≤ 0.5, low correlation
Threshold 2 = 0.5 ≤ r ≤ 0.75, medium correlation
Threshold 3 = 0.75 ≤ r ≤ 1, high correlation
Calculate and locate the correlated points in right image as well as the grid in
left image.
Calculate the parallax value

17. Experiment Results:
Destination Image showing
correlating points of different
threshold values with template size
7 x 7.
Source Image with the Grid

18. Experiment Results:
Destination Image showing
correlating points of different
threshold values with template size
of 21 x 21.
Source Image with the Grid

19. Experiment Results:
Less match points in Shadow and Homogeneous areas,
because of constant grey values
Shadow
Areas
Homogeneous
Areas

20. Conclusion:
Area based matching depends upon the quality of image as well as the template
size
In case of homogenous areas, occluded areas as well as shadow areas, points
are not well located.
With small template size, grey value uniqueness is decreased as well as the
comuptation time is increased.
With large template size, grey value uniquness is increased but so is the
geometric distortion.
Computation of parallax value gives terrain height but we were unable to
generate the surface.

21. Reference:
Schenk, T. (1999). Digital Photogrammetry. Volume I. United States Of America:
Terra Science.
Hannah, M. J., 1974. "Computer Matching of Areas in Stereo Images," Ph.D.
Thesis, Stanford University, Computer Science Department Report STAN-CS-74-
438, July, 1974.
Dr. Marsha Jo Hannah ,“Digital Stereo Image Matching Techniques”
http://www.e-perimetron.org/Vol_4_3/Balletti_Guerra.pdf
Misganu Debella-Gilo,Andreas Kääb “Sub-pixel precision image matching for
measuring surface displacements on mass movements using normalized cross-
correlation“
Misganu Debella-Gilo,Andreas Kääb “LOCALLY ADAPTIVE TEMPLATE SIZES FOR
MATCHING REPEAT IMAGES OF MASS MOVEMENTS “

22. Thank you for your
Attention