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]. Our implementation is slight different from Paper : [1] Yung-Yu Chuang, B. Curless, D. H. Salesin and R. Szeliski, "A Bayesian approach to digital matting," Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. CVPR 2001, 2001, pp. II-II [2] http://www.alphamatting.com/datasets.php
We test the performance of our algorithm using three performance metrics viz. MSE (between original image and the estimated image & alpha matte and ground truth), PSNR (between original image and the estimated image), and SSIM (between original image and the estimated image & alpha matte and ground truth).
- We didn't implemet the clustering methodology
- We experimented with size of the kernel, thus taking into account number of pixels
- Assigning weights to the kernel depending upon the size of the kernel greater than 50 x 50
Running the DEMO :
- Download 'Byaesian Matting' folder
- Run
pip install -r requirements.txt - Add the input, ground-truth and tri-map images into the input folder
- Run the 'Main.py' in the terminal
- Insert the name of the images according to the input prompt (don't need to add path, it will read the input folder's path)
- The final result will be generation of five output images that will be stored in the output folder
- The performance metrics for the input images will be displayed
Function of each '.py' file :
1] Main.py => calls all the function implemented and run
2] getKernel.py => this file constructs the kernel depending upon the set number of pixels required within the kernel.
3] kernelg.py => this file constructs the weight of the kernel
4] getFbarBbar.py => this file computes the mean 'foreground' value and mean 'background' value
5] getFcovBcov.py => this file computes the covariance of the 'foreground' and covariance of 'background'
6] matrixequation.py => this file computes the mathematical equations necesssary for the algorithm to run
7] BayesianMatting.py => this file returns the alpha matte, difference between the ground truth and the alpha matte, the estimated image, difference between estimated and original image and lastly the composited image
8] ShowNwriteIm.py => this file displays and writes the results