Algorithms and Software for High-Resolution Imaging

高分辨率图像在科学、工程、执法和医学的许多重要领域都是必不可少的。从退化图像中提取有意义信息的需求在许多领域至关重要，例如航空光学成像、监视摄影、共聚焦显微镜和医学成像。图像退化的来源因应用领域而异，但包括大气湍流、流体介质中的浊度、运动模糊和许多其他影响。图像增强、重建和恢复等图像处理任务是计算密集型的；底层物理模型需要解决数据量大且通常不适定的反问题，其中解决方案可能不连续地依赖于数据。所有这些问题都需要对算法-体系结构-应用程序交互进行重要的研究，并且可以从高性能软件系统的可用性中获益。利用问题结构和并行计算平台的迭代求解器似乎是各种大规模图像增强问题最可行的候选者。本研究项目包括以下技术课题：a)基于自适应滤波和多通道相位分集方法的有效正则化策略的盲反卷积图像恢复新迭代算法。b)利用基于Krylov子空间的多级预处理加速迭代算法。c)通过整合数值和符号计算方法部分减轻图像恢复问题病态性质的新算法。d)开发一个高性能软件包，名为“并行图像处理环境（pipe}）”，该软件包利用了本研究过程中开发的图像处理算法。

High-resolution images are essential in many important areas of science, engineering, law enforcement, and medicine. The need to extract meaningful information from degraded images is vital in many areas such as aero-optics imaging, surveillance photography, confocal microscopy, and medical imaging. Sources of image degradation vary among application areas, but include atmospheric turbulence, turbidity in a fluid medium, motion blur, and numerous other effects. Image processing tasks such as image enhancement, reconstruction, and restoration are computationally intensive; the underlying physical models require the solution to data-massive and often ill-posed inverse problems where the solution may not depend continuously on the data. All such problems require significant research into the algorithm- architecture-application interaction and can profit substantially from the availability of high-performance software systems. Iterative solvers that exploit both the structure of the problem, and the parallel computing platform, appear to be the most viable candidates for diverse, large scale image enhancement problems. This research project includes the following technical topics: a) New iterative algorithms with effective regularization strategies for blind deconvolution image restoration using adaptive filtering and multichannel phase diversity methods. b) Acceleration of the iterative algorithms utilizing Krylov subspace based multi-level preconditioning. c) New algorithms which partially mitigate the ill-posed nature of the image restoration problem by integrating numerical and symbolic computational methods. d) Development of a high performance software package, entitled "Parallel Image Processing Environment (pipe})", which utilizes the image processing algorithms developed in the course of this research.