Total variation and block-matching 3D filtering-based image reconstruction for single-shot compressed ultrafast photography

Abstract Compressed ultrafast photography (CUP), as the fastest receive-only ultrafast imaging technology by combining steak imaging and compressed sensing (CS), has shown to be a powerful tool to measure ultrafast dynamic scenes. Through a reconstruction algorithm based on CS, CUP can capture the three-dimensional image information of non-repetitive transient events with a single exposure. However, it still suffers from poor image reconstruction quality on account of the super-high data compression ratio induced by the undersampling strategy. Here, we propose a total variation (TV) combined with block matching and 3D filtering (BM3D) reconstruction algorithm to improve the image quality of CUP, which is named as the TV-BM3D algorithm. The proposed algorithm can simultaneously exploit gradient sparsity and non-local similarity for image reconstruction by incorporating TV and BM3D denoisers. Both the numerical simulations and experimental results show that, compared with the two conventional two-step iterative shrinkage/thresholding and augmented Lagrangian algorithms in CUP, the TV-BM3D algorithm can not only improve the image reconstruction quality, but also strengthen the noise immunity of this technique. It is prospected that these improvements in image reconstruction will further promote the practical applications of CUP in capturing complex physical and biological dynamics.