Title :
Fast overlapped block motion compensation with checkerboard block partitioning
Author :
Tien-Ying Kuo ; Kuo, C. C Jay
Author_Institution :
Dept. of Electr. Eng. Syst., Univ. of Southern California, Los Angeles, CA, USA
fDate :
10/1/1998 12:00:00 AM
Abstract :
The overlapped block motion compensation (OBMC) scheme provides an effective way to reduce blocking artifacts in video coding. However, in comparison with the traditional block motion compensation (BMC), its complexity of motion estimation increases significantly due to the interdependency of motion vectors. In this work, a new fast motion estimation algorithm is proposed for OBMC based on the checkerboard block partitioning and grouping in encoding. This algorithm, called the grouped OBMC (GOBMC), effectively decorrelates dependency among OBMC motion vectors and facilitates the search process. In comparison with traditional iterative OBMC motion search, GOBMC performs motion estimation only once per macroblock at the encoder, and reaches a local optimal solution with degradation of 0.05-0.1 dB. At the decoding end, we also propose a fast OBMC reconstruction scheme which reduces the complexity of multiplication to 38% of that of traditional OBMC reconstruction while preserving the same visual quality as obtained via BMC search with OBMC reconstruction
Keywords :
coding errors; computational complexity; decoding; decorrelation; image reconstruction; motion compensation; motion estimation; video coding; GOBMC; OBMC; checkerboard block partitioning; complexity; decoding; degradation; fast motion estimation algorithm; fast overlapped block motion compensation; grouped OBMC; motion estimation; motion vectors; multiplication; reconstruction scheme; search process; video coding; visual quality; Computational complexity; Cost function; Decorrelation; Encoding; Image reconstruction; Iterative algorithms; Motion compensation; Motion estimation; Partitioning algorithms; Video coding;
Journal_Title :
Circuits and Systems for Video Technology, IEEE Transactions on