Related Experiment Video
Updated: Jun 28, 2025
06:25
Time Multiplexing Super Resolving Technique for Imaging from a Moving Platform
Published on: February 12, 2014
8.5K
Non-local degradation modeling for spatially adaptive single image super-resolution
Qianyu Zhang1, Bolun Zheng1, Zongpeng Li1
1School of Automation, Hangzhou Dianzi University, Hangzhou 310018, China.
Summary
This study introduces a spatially adaptive network for blind super-resolution (SASR) that improves image quality by adaptively estimating image degradation. SASR enhances local image details, outperforming existing methods in super-resolution tasks.
Area of Science:
- Computer Vision
- Image Processing
- Artificial Intelligence
Background:
- Current single image super-resolution (SISR) methods assume uniform blur, failing on textureless areas.
- Adaptive kernel estimation is crucial for improved SISR results.
Purpose of the Study:
- To develop a spatially adaptive network for blind SISR (SASR).
- To jointly model global and local image degradation for enhanced super-resolution.
Main Methods:
- SASR utilizes an encoder-decoder architecture for global and local degradation representation.
- A cross-attention mechanism fuses these representations.
- A novel non-local spatially adaptive filtering module (SAFM) enhances local details.
Main Results:
- SASR efficiently estimates degradation and handles diverse degradation types.
- Local representations prevent single-kernel estimation issues, improving detail preservation.
- Demonstrated state-of-the-art performance in blind SISR and degradation estimation.
Conclusions:
- SASR offers superior blind SISR performance through adaptive, joint global-local degradation modeling.
- The proposed SAFM effectively preserves and enhances local image details.
- SASR achieves competitive results, addressing limitations of existing SISR techniques.
