global-local-affinity-graph

source code for the paper "A global/local affinity graph for image segmentation"

here is the code to generate the table IV for our paper

• you could also extract only part of this code easily to port single

  different graph tp generate the table I

  • to get our global/local affinity graph, you could just keep the local graph and LO graph, for example:

    W_L0 = compute_region_similarity_Sparse_penalty(feature,3,centroid,Area);
    W_GLG=assignGraphValue(W,W_L0,global_nodes);

• to combine different feature descriptor,

  • first just change the following feature=feat{k}.mlab;
    • here is the list you could change:

      
      {'mlab';

  'ch';

  'lbp';

  'siftbow100';

  'siftbow150';

  'siftbow200';

  'siftbow300'};

  
  - then you compute the affinity graph W_GLG_mlab,W_GLG_lbp, etc., 
  
  - finally, you combine them according to the fusion equation in our paper Eq.11-12

Please Note that you may not generate the exact performance listed in our paper,