2002 Fiscal Year Final Research Report Summary
The Construction of a Statistical Motion Model for Videos and its Application to Video Coding
| Project/Area Number |
12650356
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
情報通信工学
|
|---|
| Research Institution | HOKKAIDO UNIVERSITY |
|---|
Principal Investigator |
KITAJIMA Hideo Hokkaido University, School of Engineering, Professor, 大学院・工学研究科, 教授 (80001999)
|
|---|
| Project Period (FY) |
2000 – 2002
|
| Keywords | video coding / motion compensation / motion vector / interframe prediction |
|---|
| Research Abstract |
A simple, easy-to-use performance measure has been developed that can be used to evaluate motion-encoding schemes. It incorporates the average number of bits required for the residual errors after the motion-compensated inter-frame prediction using the motion vectors obtained by the motion-encoding method. Without such a performance measure one would have to actually encode the residual errors to evaluate the motion encoder. Various components involved in the encoding would then bury minute differences that exist among motion-handling schemes.
Motions in a video clip are conventionally represented by translations. The translations are assumed to be uniform over pixels in a block and vary only among different blocks. The translation model holds for movements of rigid objects, but fails, for example, for human body parts moving vigorously. A new statistical motion model has been developed. Under this model a pixel in the current frame is referred randomly to more than one pixel in the ref
… More
erence frame. The pixels in the random model are centered on a pixel determined by a translation and they are combined to define an effective prediction for the pixel in the current frame.
All objects in a video frame are not new. Some can be referred to old frames. Various possibilities for the effective use of the situation for video coding have been studied. Research is still incomplete in this aspect of videos.
The prediction of motion vectors has been investigated. The more accurate motion-vector prediction leads to the more efficient encoding of motion vectors. A prediction for a block in the current frame is conventionally formed using motion vectors taken from neighboring blocks in the current frame. It has been found that a better prediction can be formed if motion vectors from the past, reference frame are also considered in the prediction process.
Research on image segmentation has also been conducted. Rectangular blocks of pixels are currently used in motion analysis for commercial video coding systems. They will be replaced by object-based segments in an advanced image coder for the ease of video synthesis and editing. Less
|
