Spatial-Temporal Adaptive Compressed Screen Content Video Quality Enhancement

Deep learning based methods have achieved remarkable success in compressed video quality enhancement. Multiple frames are used to extract spatial-temporal information to restore high-quality frames. However, Screen Content Videos (SCV) always have dramatic content changes, it is difficult to get accurate spatial-temporal information. To address this issue, we propose a Spatial-Temporal Adaptive (STA) method. STA can extract spatial-temporal change information in SCV, and these information can guide the adaptive fusion of single-frame and multi-frame based features, thereby not only preserving strong spatial-temporal information extraction ability but also avoiding the interference caused by content abrupt changes. Specifically, we introduce a dual-branch structure for single-frame and multi-frame feature extraction in parallel. An adaptive feature fusion module is also introduced to fuse the features from two branches with the guidance of spatial-temporal change information. Experimental results demonstrate that our method outperforms previous methods in compressed SCV quality enhancement.