Presentation 2006-03-16
Semi-Supervised Object Extraction for Natural Image Matting
Weiwei DU, Kiichi URAHAMA,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Supporting strokes drawn by users called trimaps are exploited in natural image matting where an object is extracted from an image and composited against another image. In this paper, we apply a semi-supervised method for extracting a fuzzy cluster from similarity data to this task and present a method for extracting object regions by using coarse strokes by users. Rough specification of foreground or background regions instead of precise trimaps is sufficient for our method. Broad propagation window enables our method to jump gaps in objects and reduce specification strokes. However, broad window increases computational costs, hence we devise a fast algorithm for computing approximate solutions in the method.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) natural image matting / object extraction / semi-supervised cluster extraction / membership propagation
Paper # PRMU2005-254
Date of Issue

Conference Information
Committee PRMU
Conference Date 2006/3/9(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Pattern Recognition and Media Understanding (PRMU)
Language ENG
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Semi-Supervised Object Extraction for Natural Image Matting
Sub Title (in English)
Keyword(1) natural image matting
Keyword(2) object extraction
Keyword(3) semi-supervised cluster extraction
Keyword(4) membership propagation
1st Author's Name Weiwei DU
1st Author's Affiliation Faculty of Design, Kyushu University()
2nd Author's Name Kiichi URAHAMA
2nd Author's Affiliation Faculty of Design, Kyushu University
Date 2006-03-16
Paper # PRMU2005-254
Volume (vol) vol.105
Number (no) 673
Page pp.pp.-
#Pages 6
Date of Issue