目录

• grab cut
  • \(1.\) segmentation
  • \(2.\) grab cut
    • \(2.1.\) graph cut
    • \(2.2.\) code of grab cut in opencv
    • \(2.3.\) understand algorithm
    • \(2.4.\) deep grab cut

grab cut

\(1.\) segmentation

1. learning source related to interactive segmentation on github

2. project (latest achievement on CVPR2018) : Interactive Image Segmentation with Latent Diversity

3. project (latest achievement on CVPR2018) : Deep Extreme Cut: From Extreme Points to Object Segmentation

4. project (latest achievement on CVPR2019) : Curve-GCN

\(2.\) grab cut

Review of segmentation based on this blog
Learning on graph cut is based on this blog.
Learning on grab cut is based on this blog and paper.

\(2.1.\) graph cut

To help myself understand grab cut, I turn my head to learning something related to grab cut. Graph cut is one of them.

To segment a image into a object which is in our selection and the background, we can use the method of graph cut which is based on graph theory.

\(fig2.1.1\) original image is the input image as show in the left, every pixel in which can be seen as a node. Add paths to neighbor nodes, add paths to every node and s/t node, as show in the right.

As our destination is to separate object that is in our selection areas and background. In \(fig2.1.1.R\) the paths record the value of the similarity of each pixels, which is the standard of our segmentation. When the value of a path is large, it shows that it has similarity between the pixels side of the path, which is the object we are trying to preserve. Raise a method is to find the min cut of a subgraph \(fig2.1.1.R\), then separate the subgraph, we can get the objection selection.

Now, raised a issue, how to define the similarity of each pixels.

\(2.2.\) code of grab cut in opencv

\(2.3.\) understand algorithm

Based on this blog.

\(2.4.\) deep grab cut

原文地址：https://www.cnblogs.com/litun/p/12041458.html