Computational Science Technical Note CSTN-194

CSTN Home

New and Hybrid Parallel Component Labelling Algorithms and Benchmark Pattern Generators

K. A. Hawick and D. P. Playne

Archived: 2013

Abstract

Component labelling remains an important and widely used technique for analysing images and many other forms of data in various applications. Different categories of data set have different optimal component labelling algorithms. We review a number of labelling and associates algorithms and describe a series of benchmark data sets and generators for comparing component labelling algorithms. We report on extensive benchmarking tests and scalability for test patterns that can be generated on different length scales. We propose a new component labelling algorithm and also describe some ways of combining various algorithms on various parallel platforms including graphical processing units and other multi core accelerators and cpu core combinations. We also discuss the problem of simplifying other calculations such as identifying the largest component cluster or counting the number of components present.

Keywords: parallel component labelling; benchmark; scalability; GPU; CUDA; data-parallel; multi-core

Full Document Text: Not yet available.

Citation Information: BiBTeX database for CSTN Notes.

BiBTeX reference:

@TECHREPORT{CSTN-194,
        author = {K. A. Hawick and D. P. Playne},
        title = {New and Hybrid Parallel Component Labelling Algorithms and Benchmark
                Pattern Generators},
        institution = {Computer Science, Massey University, Auckland, New Zealand},
        year = {2013},
        number = {CSTN-194},
        month = {June},
        keywords = {parallel component labelling; benchmark; scalability; GPU; CUDA; data-parallel;
                multi-core},
        owner = {kahawick},
        timestamp = {2013.06.10}
}


[ CSTN Index | CSTN BiBTeX ]