Revisiting Underwater Image Enhancement for Object Detection: A Unified Quality-Detection Evaluation Framework

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Underwater images suffer from poor quality (color distortion, low contrast, reduced visibility).
• Image enhancement is commonly used, but its benefit for object detection is debated.
• Existing evaluations lack rigor and granularity.

Purpose of the Study:

• To comprehensively evaluate state-of-the-art underwater image enhancement methods for object detection.
• To propose a unified framework for rigorous, fine-grained performance assessment.
• To investigate the impact of enhancement on detection accuracy at both dataset and image levels.

Main Methods:

• Developed a unified evaluation framework integrating quality assessment (Q-index) and per-image detection (COCO mAP).
• Evaluated nine enhancement methods against modern object detectors.
• Conducted mixed-set analysis to determine theoretical performance limits of selective enhancement.

Main Results:

• Traditional quality metrics do not reliably predict object detection performance.
• Dataset-level analysis obscures significant image-level variability.
• Enhancement improves detection accuracy for low-quality inputs but can degrade performance in other cases.
• Selective enhancement yields substantial gains over original and fully enhanced datasets.

Conclusions:

• Underwater image enhancement can benefit object detection when evaluated granularly.
• Selective enhancement strategies are crucial for optimizing performance in computer vision tasks.
• This study provides evidence for the judicious application of enhancement techniques.