Fast global motion estimation based on local motion segmentation

Author

Fu, Ming Fai ; Au, Oscar ; Chan, Wng Cheong

Author_Institution

Dept. of Electr. & Electron. Eng., Hong Kong Univ. of Sci. & Technol., China

Volume

2

fYear

2003

fDate

14-17 Sept. 2003

Abstract

In MPEG-4, global motion estimation (GME) and global motion compensation can improve the visual quality of local motion estimation/compensation. However, the computational complexity of GME is very high. Even the existing fast algorithm, FFRGMET, in the MPEG-4 OM3 software is not fast enough. In this paper, we propose a fast GME algorithm called APSGME. It uses motion vectors from the PMVFAST local motion estimation to perform early termination, foreground/background block-based segmentation, feature point selection. Experimental results shown that APSGME can achieve considerable speed-up over FFRGMET while improving the visual quality simultaneously.

Keywords

feature extraction; image segmentation; motion compensation; motion estimation; video coding; APSGME; MPEG-4; Moving Picture Experts Group; PMVFAST; fast global motion estimation; feature point selection; foreground/background block-based segmentation; global motion compensation; local motion segmentation; motion vector; video compression; visual quality improvement; Cameras; Computer vision; Iterative methods; MPEG 4 Standard; Motion compensation; Motion estimation; Motion segmentation; PSNR; Pixel; Video compression;

fLanguage

English

Publisher

ieee

Conference_Titel

Image Processing, 2003. ICIP 2003. Proceedings. 2003 International Conference on

ISSN

1522-4880

Print_ISBN

0-7803-7750-8

Type

conf

DOI

10.1109/ICIP.2003.1246693

Filename

1246693