Connected Components Labeling on the GPU with Generalization to Voronoi Diagrams and Signed Distance Fields

被引：0

作者：

Rasmusson, A. ^{[1
,2
]}

Sorensen, T. S. ^{[1
]}

Ziegler, G. ^{[3
]}

机构：

[1] Aarhus Univ, DK-8000 Aarhus C, Denmark

[2] Aarhus Univ, Stereology & EM Lab, DK-8000Y Aarhus C, Denmark

[3] Nvidia Corp, Santa Clara, CA USA

来源：

ADVANCES IN VISUAL COMPUTING, ISVC 2013, PT I | 2013年 / 8033卷

关键词：

Computer vision; connected components labeling; GPU; Voronoi diagrams;

D O I：

暂无

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

Many image processing problems benefit from a complete solution to connected components labeling. This paper introduces a new data parallel labeling method based on calculation of label propagation sizes from the connectivity between pixels extracted in a pre-processing step and re-usal of established label propagation routes. The method achieves real-time performance for 2D images and it also generalizes to Voronoi diagrams and signed distance fields.

引用

页码：206 / 215

页数：10