Concurrent computation of differential morphological profiles on giga-pixel images

Authors:
Michael H. F. Wilkinson;Pierre Soille;Martino Pesaresi;Georgios K. Ouzounis
Affiliations:
Johann Bernoulli Institute, University of Groningen, The Netherlands;Geo-Spatial Information Analysis for Global Security and Stability, Institute for the Protection and Security of the Citizen, Joint Research Centre, European Commission, Ispra, Italy;Geo-Spatial Information Analysis for Global Security and Stability, Institute for the Protection and Security of the Citizen, Joint Research Centre, European Commission, Ispra, Italy;Geo-Spatial Information Analysis for Global Security and Stability, Institute for the Protection and Security of the Citizen, Joint Research Centre, European Commission, Ispra, Italy
Venue:
ISMM'11 Proceedings of the 10th international conference on Mathematical morphology and its applications to image and signal processing
Year:
2011

Citing 4
Cited 0

Computing contour trees in all dimensions

Computational Geometry: Theory and Applications - Fourth CGC workshop on computional geometry
Concurrent Computation of Attribute Filters on Shared Memory Parallel Machines

IEEE Transactions on Pattern Analysis and Machine Intelligence
Antiextensive connected operators for image and sequence processing

IEEE Transactions on Image Processing
Efficient 2-D Grayscale Morphological Transformations With Arbitrary Flat Structuring Elements

IEEE Transactions on Image Processing

Quantified Score

Hi-index	0.00

Visualization

Abstract

In this paper we provide an efficient parallel algorithm for reconstruction from markers, and multi-scale analysis through differential morphological profiles, which are top-hat scale spaces based on openings and closings by reconstruction. The new algorithms provide speed gain in two ways: (i) through parallelism, and (ii) through more efficient re-use of previously computed data. The best version of the algorithm provided a 17× speed-up on 24 cores, over computation of the same algorithm on a single core. Compared to the basic method of repeated reconstructions by a sequential algorithm, a speed gain of 25.1 times was obtained.