Toward a Unified Complementary Fusion Framework for Robust Polarimetric Imaging.

Area of Science:

• Optics and Photonics
• Computer Vision
• Image Processing

Background:

• Polarization is a key property of light, offering valuable physical cues through degree of polarization (DoP) and angle of polarization (AoP).
• Polarimetric imaging captures these parameters but faces challenges with polarizer-induced light attenuation, complicating exposure control and leading to noisy, blurry, or saturated images.

Purpose of the Study:

• To introduce PolFusion+, a unified framework for robustly generating clean and sharp polarized snapshots.
• To address challenges in polarimetric imaging, including low signal-to-noise ratio, color distortion, saturation, and motion blur.

Main Methods:

• A polarization-aware three-phase fusion scheme is employed, complementing a degraded pair of short-exposed noisy and long-exposed blurry inputs.
• Irradiance restoration phase rectifies color information from both inputs to mitigate saturation.
• Joint restoration of DoP and AoP is achieved by modeling their individual characteristics and interdependencies.
• A degradation-oriented neural network is utilized for the fusion scheme.

Main Results:

• PolFusion+ effectively mitigates saturation-induced degradation by restoring irradiance.
• Physically faithful polarization reconstruction is achieved through joint DoP and AoP restoration.
• Experimental results demonstrate state-of-the-art performance in producing clean and sharp polarized snapshots.

Conclusions:

• PolFusion+ offers a robust solution for generating high-quality polarized snapshots from degraded inputs.
• The framework's advancements in irradiance restoration and polarization reconstruction significantly benefit downstream polarimetric imaging applications.