Underwater Image Enhancement Using Adaptive Retinal Mechanisms.

Area of Science:

• Computer Vision
• Biomimetic Engineering
• Image Processing

Background:

• Underwater images suffer from significant degradation, including blurring and non-uniform color bias, hindering visual analysis and machine perception.
• Existing methods often require specific prior knowledge of underwater conditions or scene structures, limiting their applicability.

Purpose of the Study:

• To develop a self-adaptive underwater image enhancement model inspired by the teleost fish retina.
• To address blurring and non-uniform color bias in underwater imagery.

Main Methods:

• The model mimics retinal functions: horizontal cells for color bias correction, bipolar cells for edge enhancement, and ganglion cells for color correction.
• A luminance-based fusion strategy combines ON and OFF pathways for final image reconstruction.
• Global image statistics, like contrast, automatically tune filter parameters for self-adaptation.

Main Results:

• The proposed model effectively corrects non-uniform color bias and enhances edges and contrasts in underwater images.
• Evaluations demonstrate competitive performance across diverse underwater scenes.
• Significant improvements were observed in transmission map estimation and local feature point matching.

Conclusions:

• The fish retina-inspired model offers a robust solution for single-image underwater image enhancement without requiring specialized priors.
• The self-adaptive nature and biomimetic approach provide a novel and effective method for improving underwater image quality.