Reproduce “MF-LPR$^2$: Multi-Frame License Plate Image Restoration and Recognition using Optical Flow” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kihyun Na, Junseok Oh, Youngkwan Cho, Bumjin Kim, Sungmin Cho, Jinyoung Choi, Injung Kim
MF-LPR² is a framework that restores and recognizes license plates from low-quality dash cam footage by robustly aggregating information across multiple video frames. It achieves an 86.44% recognition accuracy on a novel real-world dataset, significantly surpassing prior multi-frame methods and general super-resolution models, while preserving the evidential integrity of license plate characters.
Key insights
- Robust multi-frame aggregation, enabled by advanced optical flow correction, is crucial for effectively overcoming severe real-world image degradations in LPR.
- Prioritizing content preservation over visual hallucination is essential for the forensic integrity of LPR, as generative artifacts can fundamentally alter critical information.
- Combining temporal consistency checks and spatial planar refinement significantly enhances the accuracy of optical flow estimation in highly degraded video sequences, which is vital for multi-frame alignment.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.