On 23/04/13 8:42 PM, Juan Nunez-Iglesias wrote:

This is a whole big can of worms! It really depends on what you're doing later. For example, do you want to do this hierarchically? Or do you want to do it in one shot? One shot is simplest: for each edge, keep a score, something like Pr(join(u, v)). Then, threshold this graph, ie, throw away all the edges where your probability is lower than some value. Finally, do a connected components search, and iterate over all superpixels as follows:

I think I understand what you are getting at when you say "keep a score". What you mean "hierarchically" ? Obviously I do some background reading on graph theory so I can understand how to threshold a graph a graph etc. I will doo that later tonight. Once superpixels are combined, I have no need for the graph. I was planning to compare each adjacent superpixel (i.e. each edge) and update a new segment mask as I go. But I have a few problems. Say S1 and S2 are joined. Later I compare S2 and S5 need need to be joined, I need to keep track the S2 has already be relabeled. I think I will need multiple passes, which implies new graph for each pass and repeat until no joins detected. Assuing it would work, it doesn't seem very efficient. I suspect this process to combine vertices in a graph already exists in a graph library so will investigate existing libraries. Thanks again for the ideas. Michael. --