.png){kind=icon}
.png){kind=icon}
Return
Multi-Scale Motion Alignment and Frame Reconstruction for Efficient Deep Video Compression
.png){kind=icon}
PRE
AI
DOI：10.1109/LSP.2024.3443516.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
As video data continues to grow, the burden on network transmission increases significantly. Efficient video compression techniques are crucial to meet the rising demand for multimedia content. In this letter, we propose a Multi-scale Motion Alignment and Frame Reconstruction-based Video Codec (MFVC) for efficient video compression. MFVC focuses on optimizing the motion compensation and video reconstruction processes within a deep video compression framework. First, we design a Multi-Scale Motion Alignment Network (MSMA-Net) to achieve precise motion compensation, which extracts multi-scale features from video frames and utilizes flow information for deformable convolution. Second, we design a Frame Reconstruction Network (FR-Net) to recover high-quality video frames, which utilizes reference information for feature enhancement without additional bitrate consumption. Moreover, to achieve smooth rate adjustment, we introduce a feature scaling technique. Experimental results show that MFVC reduces bitrate by 7.86%/48.34% compared to VVC (VTM 13.2) at the same PSNR/MS-SSIM.
Keywords:
Convolution
Decoding
Motion compensation
Video compression
Feature extraction
Encoding
Video codecs
Deep video compression
end-to-end video codec
flexible rate adjustment
video coding
Journal
IF:
9.6
Papers:
1.1W
Citations:
1.7W