Video encoding device, preprocessor, and video encoding method
The video encoding device and method optimize preprocessing and encoding by integrating a picture reference structure, enhancing encoding efficiency and video signal quality through neural network preprocessing and encoding.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- NIPPON HOSO KYOKAI
- Filing Date
- 2022-07-06
- Publication Date
- 2026-06-26
AI Technical Summary
Existing video encoding methods using neural networks do not effectively utilize the picture reference structure during preprocessing, leading to a potential deterioration in encoding efficiency after decoding.
A video encoding device and method that utilize a first neural network to perform preprocessing on multiple frames using a picture reference structure, optimizing the pre-processing unit and video encoding unit through a loss function to improve encoding efficiency by integrating the picture reference structure during both training and inference phases.
Enhances encoding efficiency by improving the quality of decoded video signals and maintaining encoding efficiency by utilizing the picture reference structure before encoding, thereby outperforming methods that do not use such structures.
Smart Images

Figure 0007880757000003 
Figure 0007880757000004 
Figure 0007880757000005