Section: New Results

Parallel light speed labeling: the world’s fastest connected component labeling for multicore processors

Participants : Lionel Lacassagne, Laurent Cabaret, Daniel Etiemble.

We have designed a parallel version of the Light Speed Labeling for shared-memory multicore processor. This algorithm outperforms the best algorithm by a factor x10. We are now working on the design of algorithms for GPU and manycore embedded processor and especially the TSAR architecture of LIP6 laboratory. More information is available at

• TSAR architecture: https://www-soc.lip6.fr/trac/tsar

• ALMOS operating system: https://www-soc.lip6.fr/trac/almos

• GIET-VM system: https://www-soc.lip6.fr/trac/giet-vm

The paper [20] introduces the parallel version of the Light Speed Labeling (LSL) and compares it with the parallel versions of the competitors. A benchmark shows that the parallel Light Speed Labeling is at least ×1.9 faster than all the other algo- rithms for random images. This factor reach ×3.6 for structured random images. More important, we show that thanks to its run-based processing (segments), LSL is intrinsically more efficient than all pixel-based algorithms.