Joint Denoising-Demosaicking Network for Long-Wave Infrared Division-of-Focal-Plane Polarization Images With Mixed...

Area of Science:

• Optics and Photonics
• Computer Vision
• Signal Processing

Background:

• Denoising and demosaicking are essential for processing long-wave infrared (LWIR) division-of-focal-plane (DoFP) polarization images.
• Current methods apply denoising and demosaicking sequentially, leading to cumulative errors and reduced performance.

Purpose of the Study:

• To propose a joint denoising and demosaicking method for LWIR DoFP images.
• To enhance the generalization ability of deep learning models for polarization image restoration.

Main Methods:

• A three-stage progressive deep convolutional neural network was developed for joint denoising and demosaicking.
• Mixed Poisson-Additive-Stripe noise was modeled, and a least-squares problem was used to estimate noise parameters from real LWIR DoFP images.
• Estimated noise parameters were used to generate synthetic training data for the neural network.

Main Results:

• The proposed method effectively performs joint denoising and demosaicking of LWIR DoFP polarization images.
• The noise modeling and data generation strategy improved the network's generalization ability on real-world data.
• Experimental results confirmed superior image restoration performance compared to existing sequential methods.

Conclusions:

• The joint deep learning approach offers a significant advancement in processing LWIR DoFP polarization images.
• Accurate noise modeling and data augmentation are crucial for robust deep learning-based image restoration.
• The developed method enhances the utility of LWIR polarization imaging in various applications.