Guest speaker, Dr Vitaliy Kurlin, discussing polygonal meshes of superpixels for image over-segmentation based on topological skeletons

18 February 2019

Guest speaker, Dr Vitaliy Kurlin, discussing polygonal meshes of superpixels for image over-segmentation based on topological skeletonsDr Vitaliy Kurlin presented a talk with us on the 15 February, discussing ‘Polygonal meshes of superpixels for image over-segmentation based on topological skeletons’. This talk provided students and academics with an insight into his work and how the process of superpixels could help reconstruct images with a smaller requirement of pixels, becoming more efficient while still producing a similar visual result. Throughout his talk, he explained the methods required to achieve a polygonal superpixel, studying the sequence of point cloud and Persistent-Homology features and the benefits of subpixel resolution. In His work he used ideas from Topological Data Analysis to predict major topological features of digital Images.

He finished the talk by presenting 3 original images of landscapes alongside 3 superpixel reconstructed images and asked for those present to determine which were original or not.