视频质量评价方法、装置、设备及存储介质
By extracting keyframes in the video quality evaluation system and using a ranking model and ELO rating system, the problems of insufficient samples and subjective noise in video quality evaluation are solved, and accurate comparison and stable evaluation between different time domain states are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- PENG CHENG LAB
- Filing Date
- 2023-03-24
- Publication Date
- 2026-07-17
AI Technical Summary
Existing video quality assessment systems suffer from problems such as insufficient sample size leading to limited model generalization ability, unstable absolute subjective scores, and inability to accurately compare different time-domain states.
By extracting keyframe sets from the videos to be evaluated and the training videos, a ranking model is used for ranking comparison and ranking score competitive regression. The Siamese network and fully connected layers are combined for feature extraction and dimensionality reduction. An improved K-means clustering model is used to select keyframes, and an ELO ranking system is introduced to adjust the regression order.
It reduces the impact of subjective noise, improves the accuracy and stability of video quality evaluation, reduces the need for training data, alleviates the problem of insufficient samples, and enables accurate comparison between different time domain states.
Smart Images
Figure CN116385394B_ABST