Polarized image super-resolution via a deep convolutional neural network.

Area of Science:

• Optics and Photonics
• Computer Vision
• Image Processing

Background:

• Low-resolution polarized images hinder the detailed analysis of polarization information, limiting the identification of small targets and weak signals.
• Polarization super-resolution (SR) aims to generate high-resolution polarized images from low-resolution inputs, but it is more complex than traditional intensity-based SR due to multiple channels and nonlinear cross-links.

Purpose of the Study:

• To address the challenges in polarization SR by proposing a novel deep convolutional neural network.
• To effectively reconstruct both intensity and polarization information simultaneously for improved image resolution.

Main Methods:

• Analysis of polarized image degradation processes.
• Development of a deep convolutional neural network tailored for polarization SR reconstruction.
• Design of a specialized loss function to balance intensity and polarization information restoration.

Main Results:

• The proposed deep convolutional neural network effectively performs polarization SR with a maximum scaling factor of four.
• Experimental results demonstrate superior performance compared to existing SR methods in both quantitative and visual evaluations.
• The network successfully balances the restoration of intensity and polarization information.

Conclusions:

• The developed deep learning method offers a significant advancement in polarization super-resolution.
• This technique enhances the ability to distinguish detailed polarization information and identify targets in low-resolution polarized images.
• The proposed approach provides a robust solution for reconstructing high-resolution polarized images across different degradation models.