Parallelization of Face Detection Engine

Abstract

Video processing is computationally intensive and often has accompanying real-time or super-real-time requirements. For example, video tagging and surveillance systems need to robustly analyze video and automatically recognize the faces in real time. The semiconductor industry has shifted from increasing clock speeds to a strategy of growth through increasing core counts. This shift from single core to multi-core presents a major challenge to application developers to exploit sufficient parallelism in performance-sensitive applications. This give rise to a new computation paradigm for developing more advance algorithms. In this paper, we present a method to efficiently parallelize face detection which can be extended to any object detection algorithms for SMP architectures. We also show that a well-designed parallel code of face detection algorithm will result in a performance gain in excess of 2X on dual core systems.