一橋大学
ソーシャル・データサイエンス学部・研究科
グラフィックス&ビジョン研究室 (谷田川研究室)

yatagawa2012dbgrem.jpg

Template-Based Dynamic-Background Removal for Fast Video Completion

Tatsuya Yatagawa , Yasushi Yamaguchi
The University of Tokyo
In Lecture Notes in Computer Science (Proc. of Int'l Conf. on Visual Computing 2012)

Abstract

Video completion is a technique to remove large portion of area in video by detecting other similar portion from input video. Recently, preliminarily classifying input video pixel into foreground and background is found to be efficient for accelerating video completion procedure. Though existing completion algorithms only pay attention to moving foreground, many videos contain moving background. Therefore they are originally difficult to handle videos with moving background. Many background subtraction algorithms can extract moving objects without any user guide, but they cannot distinguish moving foreground and moving background. We presents the novel approach to remove moving background region automatically and apply existing fast completion algorithms to videos with moving background.

Results

Foreground Extraction

(upper left) input video
(upper right) simple thresholding
(lower left) Zivkovic 2004 [1]
(lower right) our result

Completion 01

(left) input video
(right) output video

Completion 02

(left) input video
(right) output video