Hi Rachida

You should primarily formulate your problem clearly, firstly what are the variables, are there any constrains for them?, what is the objective function? is it composite or not if it is you should use the appropriate weights for each term. Then you should choose the appropriate minimization approach that fit your problem (are you interest on a local or global minima depending on the accuracy required, is the time consideration is a parameter, then the computer platform and its speed is important. all these points are very important.

Good Luck

Prof. S. EL-Rabaie

Hi Mr El-Rabaie, 

Thank you for your replay,

The problem is image co-segmentation so in my case the objective function follows a MRF form with three terms and they are all submodular thus th function is graph representable and we can minimize it using graph cut (global minima). 

My problem is how to construct the graph. How can i represente these three terms as edges??

Thank you again.

Open source high order CRF software such as ALE(http://www.robots.ox.ac.uk/~phst/ale.htm) and Darwin (http://drwn.anu.edu.au/) may help.

Thank you Linag Xiao