KAZE Features

>> YOUR LINK HERE: ___ http://youtube.com/watch?v=l2TkvyXavFU

This video shows some results and supplementary material from the paper: • KAZE Features. Pablo F. Alcantarilla, Adrien Bartoli and Andrew J. Davison. In European Conference on Computer Vision (ECCV), Fiorenze, Italy, October 2012. • Most multiscale 2D feature methods rely on the Gaussian scale space to detect and describer local features. However, the Gaussian kernel does not respect the boundaries of the objects, and smoothes in the same degree details and noise. In this work, we show that we can obtain much better repeatability and distinctiveness by detecting and describing features in nonlinear scale spaces, retaining important object boundaries and smoothing noise or small details. • For more information please check: • http://www.robesafe.com/personal/pabl... • The source code of the project can be downloaded from: • http://www.robesafe.com/personal/pabl...

#############################

New on site