I am working in image segmentation with super pixels. My data is a large matrix describing various attributes of each stick of pixels (such as height, width and disparity). The data comes from an image taken with a stereo camera so for each stixel I can get a respective point in 3D. From these points I have built a graph where each vertex is the stixel ID, the edges represent proximity (if a stixel is a neighbour of another) and the weights are the euclidean distance. Since I must cluster by orientation planes I thought that the branches of an MST would give me an initial set of orientation planes and starting clusters. However, I don't know how to cluster the MST's branches using the weights. I also tried graph clustering with MCL in the full graph but the results where not very good. Is there any algorithm that could cluster a tree's branches?

  • $\begingroup$ So as of now I've found two simple approaches that may help me. One is deletion of the heaviest edge, however this means I must decide on the number of clusters and deletion if the edge is inconsistent (if its weight exceeds the average of the weights of the neighbourhood. So unless I get a better suggestion I think I'll go with those. $\endgroup$ – RCountZero Nov 20 '15 at 14:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.