Computer Vision News - August 2016

Computer Vision News Trick 15 B. Ultrametric Contour Map (UCM) and C. Edge Superpixels The UCM’s Affinities matrix and the Superpixel method will give us the final edges. The UCM is an edge map with the important attribute that it produces a set of closed curves (this is not always true of the original edge map). That is to say that the UCM is a soft representation of a segmentation. For more details on UCM see " From Contours to Regions: An Empirical Evaluation " by Pablo Arbelaez et al. In the code snippet below the spAffininites compute the ultrametric contour map (UCM) given the (a) edge map and (b) the superpixel clustering of the image (the spDetect function computes the superpixels of the image). The final edges (dark black) look as follows: Nice, isn’t it? Now we've got edges between semantically meaningful regions. Let’s have another quick example with Matlab ’s famous peppers image. It is clear how the dark back edges actually separate the various peppers. Trick