GPU Accelerated Vessel Segmentation using Laplacian Eigenmaps
Lin Cheng, Hyunsu Cho, Peter A. Yoon · 2014
Laplacian eigenmap is a useful technique to improve clusterbased segmentation of multivariate images. However, this approach requires an excessive amount of computations when processing large image datasets. To that end, we present a GPU-based acceleration procedure for vessel segmentation problems. 2. Laplacian Eigenmap As described in Laskaris et. al. [1], the Laplacian Eigenmap is an effective dimensionality reduction method that maps a set of multivariate features vectors to points on the real line. Each features vector characterizes a group of neighboring pixels referred to as a patch. Once the projection is completed, a conventional method can be used to classify the projected points into one or more clusters. Generate patches • Let xi be the multi-dimensional vector representing the ith patch.