Efficient imaging through scattering media by random sampling

Hu Yifu,Jin Xin,Dai Qionghai

DOI: https://doi.org/10.1109/MMSP.2016.7813348

2016-01-01

Abstract:Imaging through scattering media is a tough task in computational imaging. A recent breakthrough technique based on speckle scanning was proposed with outstanding imaging performance. However, to achieve high imaging quality, dense sampling of the integrated intensity matrix is needed, which leads to a time-consuming scanning process. In this paper, we propose a method that exploits spatial redundancy of the integrated intensity matrix and reconstructs the complete matrix from few random samples. A reconstruction model that jointly penalizes total variation and weighted sum of nuclear norm of local patches is built with improved reconstruction quality. Experiments are performed to verify the effectiveness of the proposed method and results demonstrate that the proposed method can achieve a same imaging quality with 80% reduction of the data acquisition complexity.