| Project/Area Number |
21K17770
|
|---|
| Research Category |
Grant-in-Aid for Early-Career Scientists
|
| Allocation Type | Multi-year Fund |
|---|
| Review Section |
Basic Section 61010:Perceptual information processing-related
|
|---|
| Research Institution | Yokohama National University (2023)
Waseda University (2021-2022) |
|---|
Principal Investigator |
孫 鶴鳴 横浜国立大学, 大学院工学研究院, 准教授 (90835886)
|
|---|
| Project Period (FY) |
2021-04-01 – 2025-03-31
|
| Project Status |
Granted (Fiscal Year 2023)
|
| Budget Amount *help |
¥2,730,000 (Direct Cost: ¥2,100,000、Indirect Cost: ¥630,000)
Fiscal Year 2022: ¥1,170,000 (Direct Cost: ¥900,000、Indirect Cost: ¥270,000)
Fiscal Year 2021: ¥1,560,000 (Direct Cost: ¥1,200,000、Indirect Cost: ¥360,000)
|
|---|
| Keywords | VVC / Compression / Video coding / Video coding for machine / Transform / Filter / Intra prediction / Neural network / Low complexity |
|---|
| Outline of Research at the Start |
Emerging video compression standard Versatile Video Coding (VVC) can double the compression ratio than the previous standard at the cost of high coding complexity. This research will reduce the complexity of VVC, and further enhance its coding gain by exploiting light yet efficient neural networks.
|
| Outline of Annual Research Achievements |
This year, I mainly focused on two components in VVC. One is adaptive loop filter which is a new filter adopted in VVC. To reduce the complexity overhead of adaptive loop filter, a hardware-oriented algorithm is developed at first. In detail, filter coefficient is limited and two mapping methods are proposed. After that, the corresponding VLSI architecture is also designed, and it can realize 4K@30fps throughput. Compared with previous work, more than 41% normalized area can be saved. Besides, the proposed coefficient limitation will not influence the compression efficiency in terms of BD-rate.
The other is intra prediction which is an essential component in various VVC coding configurations. To improve the compression efficiency, several new intra coding features have been adopted in VVC. For prediction partition, QTMT is adopted while only QT was used in previous standard HEVC. For prediction mode, the number is as large as 67, while it was only 35 in HEVC. To reduce the number of partitions and modes requiring a complete rate-distortion computation process, histogram of oriented gradient information is used. As a result, compared with original VVC test model, about 69% encoding time can be saved with only 2.96% BD-rate overhead. To illustrate the universality of the proposal, the method is also implemented on VVenC and VVdeC.
Both topics are accepted by peer-reviewed international journals.
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
Within this year, I published two international journals on the topics of low-complexity VVC. One is about VVC intra prediction algorithm which is published in Journal of Visual Communication and Image Representation (Impact Factor: 2.6). The other is about VVC adaptive loop filter architecture which is published in IEEE Transactions on Circuits and Systems-II (Impact Factor: 4.4).
In addition to journal papers, I have also published some international conferences on the related topics with VVC.
|
| Strategy for Future Research Activity |
I plan to explore several potential usages by VVC in the coming next year. One direction is to utilize LLM to help improving the coding efficiency of VVC. Another direction is to use VVC to compress neural network features rather than video.
|
