Real-time minimalist imaging via spatial aberration uniformization and re-parameterized restoration

Area of Science:

• Optical Engineering
• Computer Vision
• Image Processing

Background:

• Minimalist optical systems often exhibit spatial aberration non-uniformity across fields of view (FoVs), compromising image restoration quality.
• Existing image restoration networks can be too complex for real-time applications.

Purpose of the Study:

• To present a comprehensive design framework for improving image quality and real-time restoration in minimalist optical systems.
• To address spatial aberration non-uniformity and network complexity limitations.

Main Methods:

• Developed a spatial aberration uniformization (SAU) methodology integrated with a spatial aberration uniformity metric in optical design.
• Introduced a lightweight, real-time re-parameterized image restoration network (RPIRNet) using a structural re-parameterization strategy.
• Incorporated a frequency-aware loss function to improve high-frequency component recovery.

Main Results:

• Achieved substantial improvement in imaging consistency across the full FoV.
• Demonstrated real-time inference latency of 2.15 ms with the RPIRNet.
• Validated the framework with a fabricated dual-lens optical system, showing high-quality imaging.

Conclusions:

• The proposed framework effectively improves imaging consistency and enables efficient, real-time image restoration.
• The integrated approach of optical design and network optimization shows strong potential for practical applications in advanced optical systems.