High Fidelity Single-Pixel Imaging

Single-pixel imaging (SPI) is an emerging technique which has attracted wide attention in various fields. However, restricted by the low reconstruction quality and large amount of requisite measurements, SPI's practical application is still in its infancy. Inspired by the fact that natural scenes exhibit unique degenerated structures in the low-dimensional subspace, we propose to take advantage of such local prior via convolutional sparse coding to implement high fidelity SPI. Specifically, we can represent the target scene via convolving with a set of statistically learned kernels, with the convolution coefficient matrix being sparse. We introduce the above local prior into conventional SPI framework to promote the final reconstruction quality. Experiments both on synthetic data and real captured data demonstrate that our method can achieve better reconstruction from the same measurements, or reduce the number of required measurements for the same reconstruction quality.