摘要

工作在大气媒介中的光学成像设备因大气湍流的影响不仅分辨力受到了限制，而且获得的图像存在畸变、抖动、光照不均以及模糊。在天文观测，对地遥感以及远距离监视等应用领域急需对此类退化图像进行处理以获得清晰图像。大气湍流退化图像复原正是因这些迫切需求而发展起来的一门技术，随着计算机技术，数学理论，大气物理和图像处理技术的不断发展，虽然大气湍流退化图像复原技术在理论上和实际应用中都得到了快速发展，但是由于大气湍流的随机性以及对光学成像影响的复杂性，现有的大气湍流退化图像复原技术仍然存很大的局限。所以深入研究该技术有着重要的意义。

首先依据具体的应用背景将大视场、远距离拍摄过程中因大气湍流而退化的图像作为研究对象，依据光在大气中的传输与成像相关理论，深入研究了长曝光大气湍流退化图像和短曝光大气湍流退化图像各自的退化特点和相应点的扩散函数。

其次本文通过统计分析大量的大视场、长距离条件下拍着的清晰图像的频谱特征，并通过数学方法模拟这些图像受到长曝光大气湍流退化过程的降质，然后对比分析退化前后图像频谱的变化，最终得出该类图像频谱有着大致相似的形状。基于这些固定的形状本文提出了用于简化清晰图像频谱的近似等腰三角形模型，利用该模型结合长曝光大气湍流退化图像，就能估计出点扩散函数的参数，然后得到点扩散函数，并最终完成大气湍流退化图像复原。

最后本文在提出新的长曝光大气湍流退化点扩散函数并将其应用于图像复原以后，还开发了对应的硬件系统。通过该硬件系统的开发不仅验证了本文算法能应用于实时处理系统，同时还为以后工程应用打下了一定的基础。

关键词：大气湍流，图像复原，点扩散函数估计，DSP 实现

ABSTRACT

Because of atmospheric turbulence’s effect, Optical imaging equipments which work in the atmospheric media are not only limited in resolving power, but also their images are always distorted, tilted, uneven illuminated and blurred. Degraded images from Astronomical observation, remote sensing and long range surveillance are needed for processing urgently in order to gain clear images. The technology of Atmospheric turbulence degraded image restoration is developed as for these urgent needing. With the Continuous development of computer technology, math theory, atmospheric physics, image processing, atmospheric turbulence degraded image restoration technology has been development both in theory and application on some degree, but currently they all have many limitations, so studying this technology in-depth is significant.

Firstly, according to the specific application background, the research objects are focused on the atmospheric turbulence degraded natural images which acquired under large field of view and long distance. Based on the related theory of the light transmission and imaging through atmosphere, this paper have studied the characteristic and the point spread function of long-exposure and short-exposure atmospheric turbulence degraded images.

Secondly, this paper have analyzed the spectrum feature of a large number of clear nature images acquired under large field of view and long distance, and simulated atmospheric turbulence degradation process mathematically. Through analysis, this sort of images’ spectrum has the fixed shape. Based on these fixed shape, this paper have proposed the approximate isosceles triangle principle to simplify the clear image’s spectrum. Using this principle and combining with the degraded image, this paper have estimated the point spread function successfully, and then gained the restored image.

Thirdly, this paper have introduced the hardware real-time image processing system, it was designed for verifying the proposed algorithm in this paper. This hardware system has not only verified the algorithm’s efficiency but also lay the foundation for engineering applications in the future.

摘要

目录