High-performance reconstruction method combining total variation with a video denoiser for compressed ultrafast...

Area of Science:

• Optics and Photonics
• Image Processing
• Computational Imaging

Background:

• Compressed ultrafast photography (CUP) captures high-speed dynamic events.
• Existing CUP reconstruction algorithms are slow and yield suboptimal image quality.
• Limitations hinder the practical application of CUP technology.

Purpose of the Study:

• To develop a novel, efficient, and high-quality image reconstruction algorithm for CUP.
• To address the time-consuming nature and poor imaging quality of current methods.
• To enhance the practical utility of CUP for ultrafast dynamic scene capture.

Main Methods:

• A forward mathematical model for CUP was established.
• Closed-form solutions for suboptimization problems were derived using plug-in and plug-out frames.
• An advanced neural network-based video denoising algorithm (FastDVDnet) was integrated for denoising.

Main Results:

• The proposed PnP-TV-FastDVDnet algorithm exploits spatial and temporal correlations for improved reconstruction.
• Significant improvements in Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) were observed on real CUP data.
• The method achieved comparable visual results to state-of-the-art algorithms while reducing running time by 96%.

Conclusions:

• PnP-TV-FastDVDnet offers superior image quality and significantly reduced computation time for CUP.
• The algorithm effectively overcomes the limitations of traditional reconstruction methods.
• This advancement enhances the feasibility and performance of CUP in capturing ultrafast dynamic scenes.