UPBD: Construction and Evaluation Methods of the Underwater Polarization Benchmark Dataset for Complex Scenarios
Existing underwater polarization datasets are primarily designed for deep learning training and lack diverse data types and corresponding evaluation methods for comprehensive algorithm assessment. To address this limitation, we propose the Underwater Polarization Benchmark Dataset (UPBD), which cons...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Photonics Journal |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11108231/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Existing underwater polarization datasets are primarily designed for deep learning training and lack diverse data types and corresponding evaluation methods for comprehensive algorithm assessment. To address this limitation, we propose the Underwater Polarization Benchmark Dataset (UPBD), which consists of three functionally distinct subsets designed to evaluate different aspects of underwater image restoration. The dataset is accompanied by a novel multidimensional image quality evaluation framework. Our scene variation subset evaluates polarization image restoration performance across scenes with low, high, and complex degrees of polarization. The resolution test subset quantitatively assesses spatial resolution preservation using both USAF and ISO12233 resolution test charts. The color fidelity subset provides objective color restoration analysis through a 24-color standard chart. UPBD implements rigorous acquisition protocols to ensure evaluation consistency. To demonstrate the effectiveness of our multidimensional evaluation approach, we tested six representative restoration algorithms spanning different methodologies: image enhancement algorithm, imaging model-based algorithm, three underwater polarization imaging algorithms, and a deep learning algorithm. Experimental results reveal significant performance variations across different subsets and metrics, validating the importance of comprehensive evaluation enabled by UPBD. |
|---|---|
| ISSN: | 1943-0655 |