Summary of Deep Blind Super-resolution For Satellite Video, by Yi Xiao and Qiangqiang Yuan and Qiang Zhang and Liangpei Zhang
Deep Blind Super-Resolution for Satellite Video
by Yi Xiao, Qiangqiang Yuan, Qiang Zhang, Liangpei Zhang
First submitted to arxiv on: 13 Jan 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Image and Video Processing (eess.IV)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary Recent advancements in Satellite Video Super-Resolution (SVSR) have led to remarkable progress. However, most SVSR methods assume fixed and known degradation, making them vulnerable in real-world scenes with multiple and unknown degradations. To address this issue, blind SR has emerged as a research hotspot. Existing approaches focus on blur kernel estimation but neglect another critical aspect for VSR tasks: temporal compensation, particularly compensating for blurry and smooth pixels with vital sharpness from severely degraded satellite videos. This paper proposes a practical Blind SVSR algorithm (BSVSR) that explores more sharp cues by considering pixel-wise blur levels in a coarse-to-fine manner. The algorithm employs multi-scale deformable convolution to coarsely aggregate temporal redundancy into adjacent frames using window-slid progressive fusion, followed by fine merging of adjacent features into mid-feature using deformable attention, which measures pixel blur levels and assigns weights accordingly. A pyramid spatial transformation module is also devised to adjust the solution space of sharp mid-feature, enabling flexible feature adaptation in multi-level domains. The proposed BSVSR demonstrates favorable performance against state-of-the-art non-blind and blind SR models on both simulated and real-world satellite videos. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine watching a blurry video taken from a satellite, trying to make out what’s happening. Researchers have made great progress in improving the quality of these videos, but they often assume the blur is consistent and known. In reality, this isn’t always the case. To address this issue, scientists are working on “blind” methods that can adapt to different levels of blur without knowing beforehand how much blur there will be. The proposed method, called BSVSR, uses a combination of techniques to improve the quality of blurry satellite videos by considering the amount of blur in each pixel. This allows it to prioritize the most important details and produce a clearer image. |
Keywords
» Artificial intelligence » Attention » Super resolution
