| 研究課題/領域番号 |
21K17770
|
|---|
| 研究機関 | 早稲田大学 |
|---|
研究代表者 |
孫 鶴鳴 早稲田大学, 理工学術院, 次席研究員 (90835886)
|
|---|
| 研究期間 (年度) |
2021-04-01 – 2023-03-31
|
| キーワード | Video coding / VVC / Intra prediction / Transform |
|---|
| 研究実績の概要 |
This research aims at reducing the complexity of the next-generation video compression standard Versatile Video Coding (VVC), from the aspect of both algorithm and architecture. For the algorithm, a fast intra prediction based on histogram of oriented gradient is proposed. When integrating in VVC test model, more than 50% encoding time can be saved with less than 3% coding efficiency loss. For the architecture, we propose a reconfigurable transform architecture which supports all the VVC transform types with square and rectangular sizes ranging from 4x4 to 32x32. The results show that we can reduce significant area and power consumption compared to the existing methods.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
This year, we focused on low-complexity methods of two important coding components in VVC which are intra prediction and transform. For the intra prediction, a fast algorithm is develeped. For the transform, a low-cost hardware is proposed. Both methods have been published in international conference. In addition, we have also explored deep learning technology to further improve the coding gain of VVC. By using the proposed deep learning-based in-loop filter, at most 9% coding gain can be achieved. The related results have been published in IEEE journal.
|
| 今後の研究の推進方策 |
There are two plans. First is to continue the low-complexity designs for each VVC components. We have focused on intra prediction and transform in the last year. We will focus on inter prediction, adaptive loop filter and entropy model within this year. For each component, both algorithm and architecture designs are expected. Second is to develop low-complexity neural network to improve the coding gain. Based on the proposed neural network last year, we plan to use network prune and quantization scheme to reduce the computational and memory cost of the neural network.
|
| 次年度使用額が生じた理由 |
Since there was no chance to attend the conference physically last year, the amount of budget for the traveling fee is remaining. For this remaining part, I plan to purchase a CPU server for the usage of runing VVC test model in this fiscal year. In addition, I plan to buy a GPU machine for training the neural network models.
|
