This paper judges underwater image enhancement by what it does for frame matching. It proposes two measures, local matching stability and furthest matchable frame, inside an evaluation framework built for underwater video, and finishes by showing how enhancement changes the behaviour of a complete SLAM system.
Findings
- The metric-based analysis identifies strengths and limits of existing enhancement methods and gaps in how they are usually assessed (abstract). The abstract gives no headline number.
Limits
- Abstract only so far.
Relevance
Anyone putting an enhancer such as DM_underwater in front of a mapper should ask whether it makes frames easier to match, or only nicer to look at. Compare sauder-2024-caustics and the restoration result in alvareztunon-2026-gs-underwater. See underwater perception.
Links
- arXiv
- Subfield: Field robotics, Robot learning