Learning a Non-Locally Regularized Convolutional Sparse Representation for Joint Chromatic and Polarimetric...

Area of Science:

• Optics and Photonics
• Image Processing
• Computer Vision

Background:

• Focal plane color polarization cameras are mainstream in polarimetric imaging, capturing mosaic images in one snapshot.
• Image demosaicing is crucial for these cameras but challenging due to significant data loss (15/16 pixels).
• Existing color polarization demosaicing (CPDM) methods struggle with recovering lost pixel information, leading to suboptimal results.

Purpose of the Study:

• To develop an advanced CPDM method that overcomes limitations of current techniques.
• To improve the recovery of missed pixel information in color polarization mosaic images.
• To enhance the overall quality and clarity of demosaiced polarimetric images.

Main Methods:

• A non-locally regularized convolutional sparse regularization model was proposed.
• The CPDM task was formulated as an energy function.
• Alternating Direction Method of Multipliers (ADMM) optimization was employed to solve the energy function.
• The model leverages denoising and edge-maintaining properties for improved information recall.

Main Results:

• The proposed model effectively reconstructs missed pixel information in color polarization mosaic images.
• Experimental results show superior performance compared to state-of-the-art methods on synthetic and real-world scenes.
• Quantitative measurements and visual quality assessments confirm the method's effectiveness.

Conclusions:

• The non-locally regularized convolutional sparse regularization model offers a significant advancement in CPDM.
• The method provides informative and clear results, outperforming existing techniques.
• This approach enhances the utility of focal plane color polarization cameras in polarimetric imaging.