Image Segmentation with Graph Cut

1
Image SegmentationwithGraph Cut

• Wei Wu
• Mentor Nhat Vu
• Faculty Advisor B. S. Manjunath
• Vision Research Lab

2
Image Segmentation

• Computer algorithm divides image into meaningful
  parts
• Uses quantitative analysis, content-based image
  search
• Computer-automated image segmentation is faster
  more reproducible than human calculation
• My method of image segmentation GRAPH CUT

3
Graph Cut A Real Life Application

• Problem Which distributor should deliver to each
  store?
• Assign cost of going from one store to its
  neighbor (black links) based on efficiency of
  route
• Efficiency defined by traffic, road quality, etc.
• Solutions
• Arbitrarily divide stores between the two
  distributors
• Try every possible division to find most
  efficient
• Use graph cut algorithm to eliminate least
  efficient route(Boykov Kolmogorov 2004)

Starbucks
Distributor
4
Graph Cut Applied to Images
algorithm makes min. cost cut
node for each pixel
nodes linked by edges
edges given costs
2x3 image

• My project Assign costs to edges
• Edge cost is based on degree of similarity
  between connected pixels
• Higher similarity between neighbors means higher
  cost of cutting the edge between them
• What defines similarity?
• Intensity (brightness) Texture

5
Defining Costs w/ Feature Space
(226, 5, 10)
3 x 5 image
plot features into multi-dimensional space
3 channels ? 3 features
6
Defining Costs w/ Feature Space
bigger feature distance smaller cost(vice versa)
7
Intensity-based Segmentation
8
Color-based Segmentation
red channel

time for cut 4.3 seconds
green channel

both channels
9
Training Feature Weighting
Use prior knowledge about regions to train for
which features are better for differentiating the
regions.
pixels
small overlap region feature good for
differentiating the two regions(vice versa)
pixels
B
A
pixels
10
Segmentation w/ Weighted Features
train on R G
B
A
image
train on B G
A
B
11
Conclusion

• For more complex images like retinal images, more
  features must be used for segmentation
• Graph Cut has potential to be useful for
  every-day applications
• Image cropping
• Colorization

12
Acknowledgements

• Nhat Vu
• Professor Manjunath the VRL
• Jens, Mike, Evelyn
• CNSI
• My fellow Apprentice Researchers