Contour-adaptive image restoration based on compressed sensing for ultrafast phenomena

Abstract

The compressed ultrafast photography (CUP) method is used to observe ultrafast light emission phenomena by restoring multiple images from a single observed image via a compressed sensing algorithm. However, because its regularization function is only suitable for ultrafast light emissions with lattice contours, the CUP method frequently produces artifacts in the restoration result. To solve this problem, we propose a restoration method that is suitable for ultrafast light emissions with any contour shapes. Specifically, we derive a regularization function that automatically estimates the contours of the ultrafast light emissions. Furthermore, we correct the movement of the ultrafast light emissions. By solving the inverse problem with the derived regularization function, accurate restoration results without artifacts can be obtained. Simulations using datasets that emulate fundamental phenomena show that the proposed method is superior to the conventional CUP method in terms of visual quality and the correlation with the original image.