grabcut Interactive Foreground Extraction using Iterated Graph Cuts

“grabcut”- Interactive Foreground Extraction using Iterated Graph Cuts An weizhi 2161230233 1

overview • • • Background Previous Approach Start From Graph Cut Grab Cut Conclusion 2

Background • Foreground-background segmentation 3

Previews work white brush : foreground yellow brush : boundary red brush : background 4

Previews work • Compared to the previews work ◦ Graph cut simplied the iteraction ◦ Robust for the segmentation in complex situation. 5

Start from Graph Cut background foreground 6

Start from Graph Cut • 7

Start from Graph Cut • 8

Graph Cut • Data term Smooth term 9

10

Personal understanding 11

Objective funtion • How to solve this optimization problem? Min-Cut / Max-Flow Algorithm 13

Image to Graph • Treat an image as a graph • Graph: ◦ Nodes • A background node • A foreground node • n-nodes corresponds to n-pixels ◦ Edges • Every node connect with both S and T • Every node connect with its neighbors • Treat Cut as segmentation 14

15

New Challenge How about take image into RGB colour space ? How to succeed more simple users interaction? 16

Motivation use the value histogram? Too sparse GMM (Gaussian Mixture Model) estimation 17

Colour data modelling background Background is exactly fixed Assumption: the image array z satisfied a probability distribution 18

Colour data modeling • 19

Colour data modeling • Data term Smoothness term 20

Data term • 21

• The V is unchanged from the previous term except the pixel distance calculation weight means covariance opacity GMM components 22

Method: EM algorithm • 23

Initialisation • 24

learning GMM paramaters • 26

Estimate segmentation • Repeat above steps (1)(2)(3) until convergence 27

Optimizaiton result 28

Border Matting • Define a Contour C （previous segmentation） distance centre width 29

Border Matting • Data term Smoothness term • Using DP algorithm to minimize E 30

Border Matting • Smoothness term • Data term mean covariance Gaussian probability 31

Foreground estimation • For estimate foreground pixel not from background(Bayes matte), grabcut has no blackground colours bleeding Comparing methods for border matting 32

Result 33

Result More difficult situation 34

Result 35

Failures situation • Regions of low contrast（reduce V penalty） • Camouflage, with overlap in distribution • Background material inside the user rectangle happens important to the background total distribution 36

Conclusion • Grab cut could cope with moderately difficult images with simple user interaction • It combines hard segmentation by iteration • It use border matting to make the hard boundary more smooth 37

Q&A • 38

Q&A • 39

Q&A • 40

Thank you 41