TY - JOUR
T1 - Improved parabolic prediction-based fractional search for H.264/AVC video coding
AU - Lin, Y.
AU - Wang, Y. C.
PY - 2009
Y1 - 2009
N2 - In this study, the authors propose an efficient fractional pixel search algorithm for H.264/AVC video coding to reduce the computational complexity in half/quarter-pixel motion estimation. A prediction for the optimal motion vector is derived under the assumption that the sum of absolute transform differences error surface is a symmetric parabolic function. With the optimal region, a decision rule for half-pixel search is proposed. The experimental result shows that approximately two search points are required. Based on the half-pixel search, a hierarchical quarter-pixel search is described and an extra three search points, on average, are consumed for each quarter-pixel search. Experimental results show that significant reduction in computation can be achieved, while maintaining high coding efficiency.
AB - In this study, the authors propose an efficient fractional pixel search algorithm for H.264/AVC video coding to reduce the computational complexity in half/quarter-pixel motion estimation. A prediction for the optimal motion vector is derived under the assumption that the sum of absolute transform differences error surface is a symmetric parabolic function. With the optimal region, a decision rule for half-pixel search is proposed. The experimental result shows that approximately two search points are required. Based on the half-pixel search, a hierarchical quarter-pixel search is described and an extra three search points, on average, are consumed for each quarter-pixel search. Experimental results show that significant reduction in computation can be achieved, while maintaining high coding efficiency.
UR - http://www.scopus.com/inward/record.url?scp=70349931165&partnerID=8YFLogxK
U2 - 10.1049/iet-ipr.2008.0192
DO - 10.1049/iet-ipr.2008.0192
M3 - 期刊論文
AN - SCOPUS:70349931165
SN - 1751-9659
VL - 3
SP - 261
EP - 271
JO - IET Image Processing
JF - IET Image Processing
IS - 5
ER -