Summary
International Symposium on Nonlinear Theory and its Applications
2010
Session Number:A3L-B
Session:
Number:A3L-B1
GPGPU Accelerated Scene Segmentation Using Nonparametric Clustering
Balazs Varga, Kristof Karacs,
pp.149-152
Publication Date:2010/9/5
Online ISSN:2188-5079
Summary:
This paper presents the parallel implementation of a nonparametric image segmentation method: the mean shift algorithm using joint spatial-range feature space. By considering spatial information, the mean shift can distinguish topographically differing objects in the scene, but this feature costs additional computational demand through increased number of kernel functions. The proposed algorithm runs the mode-defining kernel iterations parallel by utilizing the many-core processor architecture present in the general-purpose graphics processing unit (GPGPU). We use our own voting procedure for pixel-cluster assignment. Numerical evaluation showed that our solution efficiently speeds up the image clusterization procedure.