What is the algorithm behind photoshop's quick selection tool? [closed]

Question

I work on image processing. I just saw the quick selection tool of photoshop and I was quite impressed to see that this tool was capable to segment images along real edges, most of the time.

I could imagine two or more ways of doing what that tool does:

1. starting with an edge detector (say Canny), with adapted parameters I would just get the connected region (maybe after some dilate, then compensating with some "open" operation on the segment).

2. Doing watershed algorithm with additional boundary constraints, virtually sth like surface tension.

But maybe Im wrong.

I plan to implement a similar segmentation algorithm, so I'm interested in an idea description (like my two guesses). Can you point me to the right direction?

Heigl

Have a look at yuwing.kaist.ac.kr/courses/cs770/reading/grabcut.pdf. The algorithm looks promising and the article also mentions some other techniques, including references – Fontenot

@Oded if you look at FAQ the second point for the theme of SO questions is "a software algorithm". Can you please enlighten me why you considered this question as off-topic? – Heigl

@BarnabasSzabolcs The site of the 1st comment has vanished. Care to share your findings with a new link/anything? – Branson

My suggestion would be “GrabCut - Interactive Foreground Extraction using Iterated Graph Cuts" by Microsoft Research Cambridge, UK. That said, using a CNN trained for object recognition with a maximum selection layer and then backpropagating the resulting feature vector to the pixels is also supposed to give good results. – Portrait

github.com/opencv/opencv/blob/master/samples/cpp/grabcut.cpp – Portrait