1. Improving Image Matting using Comprehensive Sampling Sets CVPR2013 Oral

2. Outline Introduction Approach Experiments Conclusions

3. Introduction Accurate extraction of a foreground object from an image is known as alpha or digital matting.

4. Introduction Applications

5. Introduction Compositing Equation Foreground color of pixel z Observed color of pixel z Background color of pixel z Alpha value of pixel z

6. Introduction Range of α : [ 0, 1] α =1, foreground. α =0, background.

7. Introduction ill-posed problem Typically, matting approaches rely on constraints  Assumption on image statistics  User constraints like Trimap Known Foreground Known Background Unknown Region

8. Introduction Current alpha matting approaches can be categorized into 1. alpha propagation based method 2. color sampling based method

9. Introduction Alpha propagation based method  Assume that neighboring pixels are correlated under some image statistics and use their affinities to propagate alpha values of known regions toward unknown ones.

10. Introduction Color sampling based method  collect a set of known foreground and background samples to estimate alpha values of unknown pixels. The quality of the extracted matte is highly dependent on the selected samples.  missing true samples problem

11. Introduction

12. Approach Gathering comprehensive sample set Choosing candidate samples Handling overlapping color distributions Selection of best(F, B)pair Pre and Post-processing

13. Approach Gathering comprehensive sample set For each region, a two-level hierarchical clustering is applied.  first level, the samples are clustered with respect to color  second level, respect to spatial index of pixels.

14. Approach Gathering comprehensive sample set

15. Approach Choosing candidate samples Each pixel in the unknown region collects a set of candidate samples that are in the form of a foreground-background pair

16. Approach Handling overlapping color distributions

17. Approach Selection of best(F, B)pair K : chromatic distortion S : spatial statistics of the image C : color statistics

18. Approach

20. Cohen's d

21. Approach Pre-processing An unknown pixel z is considered as foreground if, for a pixel q ∈ F, TrimapExpanded Trimap

22. Approach Post-processing Eq. (2) is further refined to obtain a smooth matte by considering correlation between neighboring pixels. Cost function [5] consisting of the data term a and a confidence value f together with a smoothness term consisting of the matting Laplacian [10] [10] A. Levin, D. Lischinski, and Y. Weiss. A closed-form solution to natural image matting. IEEE Transactions on Pattern Analysis and Machine Intelligence, 30(1):228–242, 2007 [5] E. Gastal and M. Oliveira. Shared sampling for real time alpha matting. InProc. Eurographics, 2010, volume 29, pages 575–584, 2010.

23. Experiments www.alphamatting.com

24. Experiments www.alphamatting.com

25. Experiments

28. Conclusions A new sampling based image matting method New sampling strategy to build a comprehensive set of known samples. This set includes highly correlated boundary samples as well as samples inside the F and B regions to capture all color variations and solve the problem of missing true samples.