Hello everyone, i'm beginner .I'm reading a paper where after segmentation for more accurate result graph cut is used.. please guid me i have no idea about it after segmentation how it is used and and about graphcut..please give few references...my work platform is opencv + visual studio 2010 + c++. Thanks
As far as I understand, create more segments than needed, then find an optimal cut through the neighborhood graph. After that, every connected component of the graph must be united to one particular area or segment.
At the first glance, DLIB and Boost seem to provide code for min cuts.
http://dlib.net/dlib/graph_cuts/min_cut_abstract.h.html
http://www.boost.org/doc/libs/1_54_0/libs/graph/doc/boykov_kolmogorov_max_flow.html
https://en.wikipedia.org/wiki/Graph_cuts_in_computer_vision
Regards,
Joachim
The simple graph cut algorithm to be used is using in grid neighborhood i.e. (up, down, left, and right). First, you calculate unary potentials for your image which usually comes from your classifier, it can also come from your image features information.
Second, you calculate pairwise potentials which you calculate by looking at neighbor image pixels. If you have a grid neighborhood, you will only be looking for pixels in up, down, left and right direction.
Once you calculate your unary and binary (pairwise) potentials, you could use GCMEX (http://vision.ucla.edu/~brian/gcmex.html) code publicly available at UCLA website to optimize your energy potentials, and improve your segmentations.
You can read this paper by Paria et al. http://www.bmva.org/bmvc/2010/conference/paper110/paper110.pdf
for details how you can calculate unary and pairwise potentials.
All this code is in matlab, and you can write your own code to calculate energy potentials which should be given as an input to GCMex.
Please note, that this is a basic graph cut algorithm in grid neighborhood. Recent work has focused on DenseCRF settings where you consider a neighborhood where every pixel is connected to all other nodes, and one can use higher order potentials like superpixels etc. But, if you want to understand the concept and get some hands on, this energy optimization using GCMex can be a good start.
Thanks a lot sir for your nice explanations...now i'm using grabcut method in opencv
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Programming Languages