FDI-VSR: Video Super-Resolution Through Frequency-Domain Integration and Dynamic Offset Estimation.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• High-resolution imaging sensors drive demand for advanced video quality enhancement.
• Single-image super-resolution (SISR) methods applied to videos neglect temporal information, causing inconsistencies.
• Existing video super-resolution (VSR) methods often struggle with temporal coherence and global context.

Purpose of the Study:

• To develop a novel video super-resolution (VSR) framework, FDI-VSR, that integrates spatiotemporal dynamics and frequency-domain analysis.
• To improve video quality by addressing limitations of traditional SISR methods when applied to video sequences.
• To achieve superior VSR performance with reduced computational complexity.

Main Methods:

• Proposed FDI-VSR framework integrating Spatiotemporal Feature Extraction Module (STFEM) and Frequency-Spatial Integration Module (FSIM).
• STFEM utilizes dynamic offset estimation, spatial alignment, and multi-stage temporal aggregation with residual channel attention blocks (RCABs).
• FSIM transforms deep features into the frequency domain for enhanced global context capture.

Main Results:

• FDI-VSR surpasses conventional VSR methods and achieves competitive results against state-of-the-art approaches.
• Demonstrated improvements of up to 0.82 dB in PSNR on the SPMCs benchmark.
• Achieved notable reductions in visual artifacts with lower computational complexity and faster inference.

Conclusions:

• FDI-VSR effectively enhances video quality by leveraging spatiotemporal information and frequency-domain analysis.
• The proposed method offers a significant advancement in video super-resolution technology.
• FDI-VSR provides a computationally efficient and high-performance solution for VSR applications.