This project implements the Bayesian Matting technique described in Yung-Yu Chuang, Brian Curless, David H. Salesin, and Richard Szeliski. A Bayesian Approach to Digital Matting. In Proceedings of IEEE Computer Vision and Pattern Recognition (CVPR 2001), Vol. II, 264-271, December 2001[1]

The implementation details is a lttle bit differenct from Paper

1. Instead of using the continuously sliding window for neighborhood, the project apply cv2.dilation to find the neighborhood
2. Using Gaussian Mixture Model to cluster the data points not the method of Orchard and Bouman[2]
3. Without using Gaussian falloff to weights the contribution of nearby pixels

For more information see the orginal project website http://grail.cs.washington.edu/projects/digital-matting/image-matting/

The implementation was mostly adapted from Michael Rubinsteins matlab code here, http://www1.idc.ac.il/toky/CompPhoto-09/Projects/Stud_projects/Miki/index.html

More traing,testing images and different image matting Alogorithm here, http://www.alphamatting.com/index.html

origin

trimap

Done by Bayesain Matting(alpha)

composite with another landScape

[1] Yung-Yu Chuang, Brian Curless, David H. Salesin, and Richard Szeliski. A Bayesian Approach to Digital Matting. In Proceedings of IEEE Computer Vision and Pattern Recognition (CVPR 2001), Vol. II, 264-271, December 2001

[2] M. T. Orchard and C. A. Bouman. Color Quantization of Images. IEEE Transactions on Signal Processing, 39(12):2677– 2690, December 1991.