Parallelization and performance optimization of video feature extractions on multi-core systems

The low-level video feature extractions are the most time-consuming components in content-based video information retrieval systems.In this paper we study parallelization and performance optimization methods of four video feature extractions on multi-core systems.Experiments show that the processing speeds of these programs are 17 times the original processing speed on average when eight cores are used.Besides,detailed performance analysis helps us find bottlenecks and suggest ways to further improve multi-core systems performance in future.