Abstract

Topological Computer Vision is a new research area within Topological Data Analysis on the interface between algebraic topology and computational geometry. The flagship method of persistent homology quantifies topological structures hidden in unorganized data across all scales. The talk will review recent applications to Computer Vision including auto-completion of contours, parameterless skeletonisation and superpixel segmentation of images. The superpixel algorithm speeds-up higher level image processing up to 4 orders of magnitude, hence many computer vision tasks can be solved in real time. The last work is the on-going project with Microsoft Research Cambridge and is funded by the EPSRC Impact Acceleration Account through a Knowledge Transfer Secondment. The talk is based on author’s papers in Computer Vision and Pattern Recognition (2014), Computer Graphics Forum (2015), Pattern Recognition Letters (2016). All papers and C++ code are at http://kurlin.org.