The document discusses various techniques for advance image blending, including alpha blending, optimal seam calculation using minimum boundary cut and watersheds & graph cuts, and pyramid blending. It provides details on each technique, example results, and an analysis on the strengths and limitations of each approach. It also discusses considerations for real-time implementation of the techniques at 25 frames per second.

1. Advance Image Blending
Presenters:
Muhammad Naeem Tayyab
Muhammad Umer Kakli

2. Outline
Alpha Blending
Optimal Seam Calculation
- Minimum Boundary Cut
- Watersheds & Graph Cuts
Pyramid Blending
Questions/Suggestions

3. Alpha Blending
Each image is multiplied by a weighting function
which decreases or increases monotonically in
overlapped area; the resulting images are then
summed to form the mosaic.

4. Results of Alpha Blending

5. Results of Alpha Blending

6. Results of Alpha Blending

7. Results of Alpha Blending

8. How to find an Optimal Cut?
Optimal seam finding methods try to place the
seam where the intensity difference between the
two images is as low as possible.

9. Similarity Measures
Following are the two similarity measures
for all values of i,j
e(i,j)=255-abs(I1(i,j)-I2(i,j))
e(i,j)=(I1(i,j)-I2(i,j))^2
e(i,j)=abs(I1(i,j)-I2(i,j))/max(I1(i,j)-I2(i,j)))

10. Negative of the Difference

11. Davis Similarity Measure

12. Minimum Error Boundary Cut
The minimal cost path through the error surface is
computed from the following equation
The minimum cut was the optimal seam between
the two images.

13. Optimal Seam

14. Results

15. Results

16. Results

17. Results

18. Optimal Seam using Watersheds
Watersheds
- A labeled matrix identifying the watershed regions of the
input matrix, which can have any dimension.
We Apply watershed on Negative of Difference
image.
Optimal Seam is calculated using the same
method discussed above.

19. Results (Watershed Region)

20. Optimal Seam

21. Results

22. Results

23. Results

24. Results

25. Pyramid Blending
1
0
1
0
1
0
Left pyramid blend Right pyramid
http://graphics.cs.cmu.edu/courses/15-463/2010_spring/Lectures/blending.pdf

26. Pyramid Blending
 General Approach:
1. Build Laplacian pyramids LA and LB from images A and B
2. Build a Gaussian pyramid GR from selected region R
3. Form a combined pyramid LS from LA and LB using nodes
of GR as weights:
• LS(i,j) = GR(I,j,)*LA(I,j) + (1-GR(I,j))*LB(I,j)
4. Collapse the LS pyramid to get the final blended image
http://graphics.cs.cmu.edu/courses/15-463/2010_spring/Lectures/blending.pdf

27. Results

28. Results

29. Results

30. Results

31. Analysis
 Alpha blending is good as far as there is no moving objects in the
images.
 The results of Pyramid blending is reasonably good (5 level
Pyramid reduction and expansion was used in that) to remove the
parallax problem. For Videos, ghosting effects can create a
problem.
 The results with Optimal Seam Finding methods are very good but
there is a visible seam as linear blending was not performed after
finding the optimal seam. So, after apply some blending
techniques to remove the intensity difference at seam will results
the best.

32. Real Time Implementation (25 FPS)
For offline processing you can implement any of
the blending techniques, as far as online
processing is concern the implementations
contain nested for loops and multiple if
statements, if one can remove the loops by
finding the generic mathematical relation, then
might be these techniques are suitable for
requirement of 25 frames per second.

33. References
 Arturo F., Serge B., “Removing pedestrians from Google Street
View images”, IEEE International Workshop on Mobile Vision
(IWMV), June 2010.
 A. Efros, W. Freeman. “Image quilting for texture synthesis and
transfer”, In Proceedings of SIGGRAPH 2001.
 N. Patrik, “Image Stitching using Watersheds and Graph Cuts”,
Center of Mathematical Sciences, Lund University, Sweden.

34. Q&A