Accurate Foreground Extraction using Hierarchical Background Subtraction

Abstract

In this paper, we propose an accurate foreground extraction algorithm using hierarchical back-ground subtraction. We apply three different background subtraction methods to 3-level image pyramid. At the top image, spatial correlation-based background subtraction is applied and foreground candidates are extracted. Edge-weighted background subtraction is then applied to the foreground candidates in the middle image to prevent misclassifying a foreground pixel which has similar intensity with background. Finally, at the bottom image, color correlation-based background subtraction.