Single-mask sphere-packing with implicit neural representation reconstruction for ultrahigh-speed imaging

Area of Science:

• Optical imaging
• High-speed photography
• Computational imaging

Background:

• Single-shot, high-speed 2D optical imaging is crucial for analyzing transient phenomena.
• Compressed optical-streaking ultra-high-speed photography (COSUP) captures events at 1.5 million frames per second using coded apertures and galvanometer scanners.
• Randomly coded apertures in COSUP lead to complex reconstruction and video artifacts.

Purpose of the Study:

• To design a non-multiplexing coded aperture for COSUP.
• To develop an implicit neural representation for efficient video reconstruction.
• To improve the speed and quality of ultra-high-speed video capture.

Main Methods:

• Designed a non-multiplexing coded aperture for COSUP utilizing sphere packing (SP) properties.
• Implemented a galvanometer scanner for linear scanning and SP encoding patterns for uniform space-time sampling.
• Developed a self-trainable implicit neural representation for single-measurement reconstruction.

Main Results:

• The proposed non-multiplexing coded aperture simplifies reconstruction and reduces artifacts compared to random apertures.
• The implicit neural representation significantly reduces training time and eliminates the need for training datasets.
• The new scheme enables the reconstruction of more ultra-high-speed frames from a single measurement.

Conclusions:

• The novel COSUP system with a non-multiplexing coded aperture and implicit neural representation offers enhanced performance for ultra-high-speed imaging.
• This approach overcomes limitations of previous COSUP methods, enabling more effective study of transient events.
• Simulations and experiments validate the effectiveness of the proposed encoding and reconstruction scheme.