Title:	Localized search for high definition video completion
Authors:	Benoit, Jocelyn Paquette, Eric
Citation:	Journal of WSCG. 2015, vol. 23, no. 1, p. 45-54.
Issue Date:	2015
Publisher:	Václav Skala - UNION Agency
Document type:	článek article
URI:	http://wscg.zcu.cz/WSCG2015/!_2015_Journal_WSCG-No-1.pdf http://hdl.handle.net/11025/17138
ISSN:	1213–6972 (hardcopy) 1213–6980 (CD-ROM) 1213–6964 (online)
Keywords:	kompletace videa;vysoké rozlišení;odstranění objektu;koherence patchesů;lokalizované vyhledávání;multirozlišení
Keywords in different language:	video completion;high resolution;object removal;patches coherence;localized search;multiresolution
Abstract in different language:	This paper presents a new approach for video completion of high-resolution video sequences. Current state-ofthe- art exemplar-based methods that use non-parametric patch sampling work well and provide good results for low-resolution video sequences. Unfortunately, because of memory consumption problems and long computation times, these methods handle only relatively low-resolution video sequences. This paper presents a video completion method that can handle much higher resolutions than previous ones. First, to address the problem of long computation times, a dual inpainting-sampling filling-order completion method is proposed. The quality of our results is then significantly improved by a second innovation introducing a coherence-based matches refinement that conducts intelligent and localized searches without relying on approximate searches or compressed data. Finally, with respect to the computation times and memory problems that prevent high-resolution video completion, the third innovation is a new localized search completion approach, which also uses uncompressed data and an exact search. Combined together, these three innovations make it possible to complete high-resolution video sequences, thus leading to a significant increase in resolution as compared to previous works.
Rights:	© Václav Skala - UNION Agency
Appears in Collections:	Volume 23, Number 1 (2015)

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.