光是物理世界视觉信息的重要载体，光信号的高分辨率采集成像在从宏观到微观的各个领域具有广泛应用和重要意义。然而，在拍摄动态场景时，由低曝光产生的测量噪声和由物体运动造成的图像畸变均会降低成像分辨率。另外，光路中的背景杂散光和介质散射也会造成分辨率的退化。因此传统成像方法无法获得理想的成像结果。本文通过光学、信号处理以及统计学习等学科交叉，针对动态和抗散射的高分辨率成像进行了系统的研究，以计算摄像原理为核心，在成像过程中加入计算，耦合感知高维连续光信号，解耦重建动态和抗散射的高分辨图像，突破传统光学成像分辨率退化的瓶颈。本文的主要创新点如下： 1. 针对动态成像中因低曝光产生的测量噪声降低分辨率的问题，构建了噪声阈值约束和截断梯度算子，建立了光子时序的泊松最大似然耦合模型，提出了复数域 Wirtinger 优化联合重建算法，将变量空间从幂次增长降低为线性增长，有效去除测量噪声，提高分辨率，并使图像曝光时间降低了一个数量级。 2. 针对动态成像中因物体运动产生的图像畸变降低分辨率的问题，刻画了空间频谱径向连续性和方向性分布规律，提出了场景自适应稀疏采样的计算光照方法和运动畸变补偿重建算法，在提高分辨率的同时进一步将采集数据量减少约 60$\%$，从而解决了亿级像素百纳米级高分辨率的亚秒级动态成像难题。 3. 针对光路中的背景杂散光降低分辨率的问题，构建了结构光片计算成像系统，设计样本旋转以突破光线在传播方向无法编码的瓶颈，使造成分辨率不均的单向结构光片变为多向，并提出光学传递函数校正算法，有效去除光路杂散光和系统畸变，同时实现了各向同性的 2 倍超分辨率深度层析成像。 4. 针对光信号受介质散射畸变而降低分辨率的问题，构建了单像素抗散射计算成像系统，使用结构光片照明和单像素探测器耦合采集场景信息，从而避免了散射畸变对采集数据的影响，并从一维采集数据中解耦多维图像，突破了全光路因介质散射难以清晰成像的瓶颈，实现了毫米级深度抗散射的高分辨率成像。

Light is an important carrier of visual information in the physical world. High-resolution acquisition of light signals has wide applications and great significance in various fields from macrocosm to microcosmos. However, when imaging a dynamic scene, both the measurement noise caused by low exposure and the image distortion introduced by object motion degrade imaging resolution. In addition, background stray light and medium scattering also cause severe degeneration of imaging resolution. Therefore, it is difficult for conventional imaging methods to obtain satisfying results. In this dissertation, following the principle of computational imaging, we focus on the research of dynamic and anti-scattering high-resolution computational imaging, which is across the disciplines of optics, signal processing and statistical learning. Specifically, computation is introduced into the imaging process, by which high-dimensional light signals are coupled and captured, and high-resolution images free from noise and aberration are finally decoupled and reconstructed to break the degeneration limit of imaging resolution.The main contributions of the dissertation include:1. Aiming at the problem that the resolution of dynamic imaging is degraded by measurement noise due to low exposure, we introduced a noise threshold constraint and a truncated gradient solver, established a Poisson maximum likelihood model describing the random arrival sequence of photons at the detector, and proposed a complex-field joint reconstruction algorithm using the Wirtinger derivatives. The method reduces the increase scale of variable space from exponential to linear, and is effective to remove measurement noise and enhance resolution with an order-of-magnitude reduce of exposure time. 2. Aiming at the problem that the resolution of dynamic imaging is degenerated by reconstruction distortion due to sample motion, we studied the statistical distribution of natural images' spatial spectra, and proposed a content-adaptive computational illumination method for sparse sampling of the scene's spatial spectrum, and a motion-corrected reconstruction algorithm to correct for the distortion. The resolution is effectively enhanced with the required number of shots being reduced by 60%. Consequently, we realized sub-micron high-resolution imaging with a sub-second imaging speed and a wide view field at the hundred-million-pixel scale. 3. Aiming at the problem that the imaging resolution is degraded by background stray light, we established the structured light-sheet computational imaging system, in which sample rotation is introduced to break the limit of light modulation in the propagation path, which realizes multi-direction modulated light sheet that solves the anisotropic resolution problem. We also proposed an optical-transfer-function corrected algorithm that is effective to remove stray light and system aberration, and simultaneously realizes 2× isotropic super resolution for deep-tissue tomography. 4. Aiming at the problem that the imaging resolution is degenerated by medium scattering, we established a single-pixel anti-scattering computational imaging system, which utilizes the proposed multi-direction structured light-sheet illumination and a single-pixel detector to decouple the scene's information, and retrieves two-dimensional images from the one-dimensional captured data. The system is effective to avoid negative influences of scattering distortion on captured data, and is able to break the limit of high-resolution imaging through scattering medium. Consequently, we realized deep-tissue anti-scatting imaging with a millimetre-level imaging depth.