本文是《MRF GraphCut系列》系列的一部分：
- Markov Random Field (MRF) and Graph-Cut (1)
- Markov Random Field (MRF) and Graph-Cut (2)
- Markov Random Field (MRF) and Graph-Cut (3)（本文）
Implemented Loopy Belief Propagation [wiki], which is a more general optimization approach for Markov Random Field [post 1] [post 2]. Different from Graph Cut, it can be extended easily for non-grid graphs and non-binary cases. Here are some experiment results on binary and gray-scale image restoration.
Some experiment results:
Updates:
Another interesting direction for image denoising is convex optimization, such as total-variation minimization. A more detailed discussion can be found here.
Comments