In this manuscript, we propose a novel framework of computational ghost imaging, i.e., ghost imaging using deep learning (GIDL). With a set of images reconstructed using traditional GI and the corresponding ground-truth counterparts, a deep neural network was trained so that it can learn the sensing model and increase the quality image reconstruction. Moreover, detailed comparisons between the image reconstructed using deep learning and compressive sensing shows that the proposed GIDL has a much better performance in extremely low sampling rate. Numerical simulations and optical experiments were carried out for the demonstration of the proposed GIDL.

Recent trends in three-dimensional (3D) display technologies are very interesting in that both old- fashioned and up-to-date technologies are being actively investigated together. The release of the first commercially successful 3D display product raised new research topics in stereoscopic display. Autostereoscopic display renders a ray field of a 3D image, whereas holography replicates a wave field of it. Many investigations have been conducted on the next candidates for commercial products to resolve existing limitations. Up-to-date see-through 3D display is a concept close to the ultimate goal of presenting seamless virtual images. Although it is still far from practical use, many efforts have been made to resolve issues such as occlusion problems.

Computational imaging systems algorithmically post-process acquisition images either to reveal physical quantities of interest or to increase image quality, e.g., deblurring. Designing a computational imaging system requires co-design of optics and algorithms, and recently Deep Lens systems have been proposed in which both components are end-to-end designed using data-driven end-to-end training. However, progress on this exciting concept has so far been hampered by the lack of differentiable forward simulations for complex optical design spaces. Here, we introduce dO (DiffOptics) to provide derivative insights into the design pipeline to chain variable parameters and their gradients to an error metric through differential ray tracing. However, straightforward back-propagation of many millions of rays requires unaffordable device memory, and is not resolved by prior works. dO alleviates this issue using two customized memory-efficient techniques: differentiable ray-surface intersection and adjoint back-propagation. Broad application examples demonstrate the versatility and flexibility of dO , including classical lens designs in asphere, double-Gauss, and freeform, reverse engineering for metrology, and joint designs of optics-network in computational imaging applications. We believe dO enables a radically new approach to computational imaging system designs and relevant research domains.

In the conventional weighted Gerchberg-Saxton (GS) algorithm, the feedback is used to accelerate the convergence. However, it will lead to the iteration divergence. To solve this issue, an adaptive weighted GS algorithm is proposed in this paper. By replacing the conventional feedback with our designed feedback, the convergence can be ensured in the proposed method. Compared with the traditional GS iteration method, the proposed method improves the peak signal-noise ratio of the reconstructed image with 4.8 dB on average. Moreover, an approximate quadratic phase is proposed to suppress the artifacts in optical reconstruction. Therefore, a high-quality image can be reconstructed without the artifacts in our designed Argument Reality device. Both numerical simulations and optical experiments have validated the effectiveness of the proposed method.

Fourier ptychographic microscopy (FPM) is a newly developed computational imaging technique that can provide gigapixel images with both high resolution (HR) and wide field of view (FOV). However, there are two possible reasons for position misalignment, which induce a degradation of the reconstructed image. The first one is the position misalignment of the LED array, which can largely be eliminated during the experimental system building process. The more important one is the segment-dependent position misalignment. Note that, this segment-dependent positional misalignment still exists, even after we correct the central coordinates of every small segment. In this paper, we carefully analyze this segment-dependent misalignment and find that this global shift matters more, compared with the rotational misalignments. According to this fact, we propose a robust and fast method to correct the two factors of position misalignment of the FPM, termed as misalignment correction for the FPM misalignment correction (mcFPM). Although different regions in the FOV have different sensitivities to the position misalignment, the experimental results show that the mcFPM is robust with respect to the elimination of each region. Compared with the state-of-the-art methods, the mcFPM is much faster.

Gabor holography is an amazingly simple and effective approach for three-dimensional (3D) imaging. However, it suffers from a DC term, twin-image entanglement, and defocus noise. The conventional approach for solving this problem is either using an off-axis setup, or compressive holography. The former sacrifices simplicity, and the latter is computationally demanding and time-consuming. To cope with this problem, we propose a model-based holographic network (MB-HoloNet) for three-dimensional particle imaging. The free-space point spread function (PSF), which is essential for hologram reconstruction, is used as a prior in the MB-HoloNet. All parameters are learned in an end-to-end fashion. The physical prior makes the network efficient and stable for both localization and 3D particle size reconstructions.