Joint Video Team (JVT) of ISO/IEC MPEG & ITU-T VCEG
(ISO/IEC JTC1/SC29/WG11 and ITU-T SG16 Q.6)
7
th
Meeting: Pattaya II, Thailand, 7-14 March, 2003
Document: JVT-G016
Filename: JVT-G016.doc
Title: Fast Motion Estimation for JVT
Status: Input Document to JVT
Purpose: Proposal
Author(s) or
Contact(s):
Zhibo Chen, Peng Zhou, Yun He
Dept. of Electronic Engineering, Tsinghua
University, Beijing, China
GuoZhong Wang, SVA, Shanghai
Tel:
Email:
86-10-62781413
chenzhibo@tsinghua.org.cn
hey@tsinghua.edu.cn
Source:
Tsinghua University, China and SVA Group, China
_____________________________
1. Introduction
In Awaji meeting, we proposed a hybrid Unsymmetrical-cross Multi-Hexagon-grid Search
(UMHexagonS) algorithm for integer pel motion estimation together with a Center-Biased Fractional Pel
Search (CBFPS) algorithm for JVT [1].
The proposed algorithm proposed in [1] shows very good capability in keeping the rate distortion
performance (maximum distortion less than 0.1 dB) for different sequences from QCIF format to HD
(High Definition) format with different motion degree, as well as a great computation reduction up to
90% compare to Fast Full Search used in reference software can be achieved.
And in this proposal we refined two parts in UMHexagonS algorithm proposed in [1]: initial search point
prediction and early termination. By refinement on these two parts, averagely more than 90% (up to
95.54%) of computation reduction and averagely 0.04dB (maximum less than 0.096dB) PSNR drop
compared with that of Fast Full search algorithm adopted in JM reference software. Another advantage is
that a tradeoff between search speed and reconstructed quality can be achieved by purposefully change a
modulation factor in out algorithm.
We will describe the refined two parts in the following sections: initial search point prediction and early
termination. The main body of the UMHexagonS algorithm is the same that described in [1][2].
2. Initial search point prediction
Initial search point prediction is an important technique introduced by many fast motion estimation
algorithms [3,4] setting the search area around the MBD (Minimum Block Distortion) point of the whole
search window in order to improve the performance of motion estimation. Median prediction as described
in [6] is frequently used in many algorithms and standard. Motion vectors of the collocated block in the
previous frame and of the spatially adjacent blocks are also used in [3] as initial search point predictors.
According to the multiple reference frames and multiple block modes adopted in JVT, we proposed four
kinds of prediction modes in this proposal:
a) Median Prediction
As Fig.1 shows, median predictor is used in median prediction of motion vectors, the median value of the
adjacent blocks on the left, top, and top-right (or top-left) of the current block is used to predict the
motion vector of the current block (as Fig.1 shows):
The predicted motion vector is . Some rules specify the predicted motion
vector value has been defined in [6]: when block A lies outside the picture or GOB(Group of Blocks)
boundary, it is replaced by (0,0), when block C lies outside the picture or GOB boundary, it is replaced by
File:raw.doc Page: 1 Date Saved: 2009-09-23
剩余11页未读，继续阅读
资源评论
