研究課題/領域番号 |
21K17770
|
研究種目 |
若手研究
|
配分区分 | 基金 |
審査区分 |
小区分61010:知覚情報処理関連
|
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 – 2024-03-31
|
研究課題ステータス |
交付 (2022年度)
|
配分額 *注記 |
2,730千円 (直接経費: 2,100千円、間接経費: 630千円)
2022年度: 1,170千円 (直接経費: 900千円、間接経費: 270千円)
2021年度: 1,560千円 (直接経費: 1,200千円、間接経費: 360千円)
|
キーワード | Video coding / VVC / Video coding for machine / Transform / Filter / Intra prediction / Neural network / Low complexity |
研究開始時の研究の概要 |
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.
|
研究実績の概要 |
This research aims at reducing the complexity of the latest video compression standard Versatile Video Coding (VVC), from the aspect of both algorithm and architecture. For the algorithm, we developed a fast intra encoding for video coding for machine (VCM) framework, and a low complexity convolutional neural network filter. For the architecture, we proposed a reconfigurable multiple transform selection architecture for VVC.
|
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 proposed a fast VVC intra coding method which aims at reducing machine-to-machine connection in the IoT society. The results have been presented in the CAS-society flagship conference ISCAS. In addition, we also developed a reconfigurable architecture for VVC transform. The related results are published in IEEE Transactions on Very Large Scale Integration Systems. Regarding the neural network extension beyond VVC, we developed a low-complexity filter and the results are published in the journal of IEEE Multimedia. We also present several other international conferences.
|
今後の研究の推進方策 |
There are two targets. For the algorithm, we plan to continue the work on fast intra coding for machine. Currently, we only attempted the object detection and instance segmentation. We will evaluate and analyze more machine vision tasks in the future. For the architecture, we aim at designing an area-efficient architecture for VVC adaptive loop filter.
|