1. Interactive Image Cutout- Lazy Snapping
Hu Junfeng
“Lazy Snapping”, SIGGRAPH 2004
Yin Li, Jian Sun, Chi-Keung Tang, Heung-Yeung Shum

2. Interactive image cutout
Lazy snapping
Demo
Grabcut
Image cutout is the technique of removing an object from its background

3. Interactive image cutout
Lazy snapping
Demo
Grabcut
Image cutout is the technique of removing an object from its background

4. Lazy snapping Step 1: a quick object marking step
Work at a coarse scale
Specifies the object of interest by a few marking lines
Step 2: a simple boundary editing step
Work at a finer scale
Edit the object boundary by simply clicking and dragging polygon vertices

5. Object marking UI design Representative clustering centers
Two groups of lines for the representative parts of foreground and background
Representative clustering centers
K-means method to obtain 64 clusters
for each class
: for foreground
: for background

6. K-means clustering Iterating the 4 steps below Seed initialization
Assigning elements
Seed updating
Assigning again

7. Object marking Foreground/background image segmentation
A typical graph-cut problem
Intuition:
classifying the pixels into two groups, which has the Similar feature in this group;
each group has the smoothness assumption, a Commonly used prior knowledge

8. Graph cut image segmentation
An image cutout problem can be posed as a binary labelling problem on a graph G=(V, E)
V: the nodes represent all the pixels
E: the edge linking two neighboring pixels (4-neighborhood)
i: the i-th node
Background
Foreground
Edge

9. Graph cut image segmentation
Corresponding to above 2 intuitive steps
Define the likelihood energy :
Define the prior energy :
Minimize the above two terms simultaneously
Encoding the cost when the label of node i is xi
The smaller, the better
Encoding the cost when the label of node i and node j is xi and xj
The smaller, the better

10. Graph cut image segmentation
The likelihood energy
The prior energy

11. Graph cuts
Min cut == Max flow

12. Max flow problem Bottleneck problem
General algorithms: Ford-Fulkerson algorithm, push-relabel maximum flow
new algorithm by Boykov, etc

13. Boundary editing Boundary as editable polygon UI design/Tools
First vertex – border pixel with highest curvature
Next vertices: furthest boundary pixel from previous polygon
Stop when distance is below some threshold
UI design/Tools
Direct vertex editing
Overriding brush
Using graph cuts

14. Experimental results

15. 分组大作业 Project 1 彩色直方图均衡优化 1人组 时间：12月11号 Project 2 图像分割 2人组 提交时间：12月11号