Graph cuts learning

Want to put the picture first , This is drawn after studying , Although the effect is not very good .

Here is how I transformed the original image into HSV and RGB passageway , Then we did it for each single channel Graph Cuts.

1.Graph Cuts It is a method of image segmentation , This method solves the energy function by using the form of graph . Determining the energy function can give corresponding weights to the edges of the graph , Thus, the energy function can be completely converted into a S/T chart , In using the properties of graph to solve the maximum flow of graph / Minimum cut , Find the global optimum, that is, the global energy is minimum .

2.graph cut Is based on graph( graph theory ) Various segmentation algorithms , Such as NCut, Minimum cut ,RatioCut, Graph cut, etc ,graph cuts Is graph cut , Used to solve problems based on MRF A method of the energy equation of .

shortcoming ：
The result of the second kind of segmentation is that the global energy function is the smallest , Global optimal solution , Multi class problems cannot be optimized , It can only be a local optimal solution ;
Processing images with obvious differences in pixel values has advantages , High contrast images ,
The segmentation effect is poor when the processed image contains noise or occlusion , Easy to get wrong segmentation ;
You need to manually mark some front and rear pixels , There is human intervention ;
Initialize the foreground and background pixels Different points will lead to different degrees of the final segmentation effect , graphcuts The algorithm is very sensitive to user interaction ;
When the image contains complex background , More user interaction is needed , Use the brush / A brush Mark more backgrounds And foreground pixels .

The code download ：GitHub - DamonZCR/GraphCut: Use the maximum flow and minimum cut to realize image segmentation