Minimalist and High-Quality Panoramic Imaging With PSF-Aware Transformers.

Area of Science:

• Computer Vision
• Computational Imaging
• Optics

Background:

• High-quality 360° panoramic images are crucial for computer vision.
• Conventional systems are bulky, limiting use in mobile and wearable applications.
• Minimalist imaging systems are desired for portability.

Purpose of the Study:

• To develop a Panoramic Computational Imaging Engine (PCIE) for minimalist and high-quality panoramic imaging.
• To address image quality degradation in compact panoramic imaging prototypes.
• To introduce a novel approach leveraging optical system priors for image recovery.

Main Methods:

• A Minimalist Panoramic Imaging Prototype (MPIP) based on Panoramic Annular Lens (PAL) design was developed.
• Two processing pipelines, Aberration Correction (AC) and Super-Resolution and Aberration Correction (SR&AC), were proposed.
• A Point Spread Function (PSF) representation method and a PSF-aware Aberration-image Recovery Transformer (PART) network were introduced.
• The PALHQ dataset was created for training and evaluating real-world panoramic imaging.

Main Results:

• The PCIE achieved impressive imaging results on synthetic and real-world benchmarks.
• The PSF representation and PART network effectively corrected aberrations and improved image quality.
• The PALHQ dataset facilitates research in low-level vision for panoramic imaging.

Conclusions:

• The proposed PCIE offers a viable solution for minimalist and high-quality panoramic imaging.
• The PSF-aware PART network demonstrates strong performance in image recovery tasks.
• The study provides valuable insights and resources for advancing compact panoramic imaging technology.