A New Real-Time Foreground Extraction Method using Cubic Codebook and Graph-Cut Algorithm

Seungjong No, Yeungkon Woo, and Moongu Jeon

Keywords

Foreground Extraction, Cubic-Codebook, Graph-Cut Method

Abstract

This paper presents a new real-time foreground extraction method for fixed camera. Most of existing methods cannot ensure reliable moving objects segmentation performance because they only consider temporal and local region information of pixels in a scene. To achieve more accuracy detection, we develop global spatio-temporal background subtraction method based on new cubic codebook and graph-cut algorithm. The cubic codebook is designed for efficient background modeling and fast image-graph construction. Also, graph-cut method is employed for optimal foreground segmentation as combining the spatial context of input frame and the modeled temporal context by our codebook. From computer simulation, we can achieve more accuracy than references.

Important Links:



Go Back