随机介质中成像模型和图像恢复方法研究

Imaging Equipment not only records the information of objects in the scene, but also transmission media. in essence, information in image includes only Geometric orientation and radiance distribution of the observed objects. Measured at the wavelength of light, most of material in light way is rough. The material modulates light. When the nature of the medium is changed, imaging processes become more complex. Some phenomenon, such as target position drift, shape distortion and brightness fluctuations, can be found. They reduce the quality of observation data in application. This project studies the imaging principle in the random medium, explores the image degradation law, and restores the image. The goal is to restore the diffraction limited resolution image. First, conjugate relationship mechanism is researched, and geometry relationship between object-image is described. Then, Degradation of radiation energy distribution mechanism in random medium is researched, and domain stability condition is proposed. Last, the image restoration methods are researched based on the two models. The research in this project is one of important parts in image sequence processing research and its application. The achievement by this project is hoped to promote the development of machine vision and adaptive signal processing.

光学成像设备不但记录场景中物体的信息，也记录光传输介质的信息。从本质上讲，图像反映的信息只包括观测对象的几何方位和辐射分布信息。以光的波长来衡量，光路上的大多数物质都是粗糙的，这些物质对光进行调制。当介质的性质随机变化时，成像过程变得更复杂，表现为目标位置漂移，形状畸变和亮度涨落。这些现象在应用中降低了观测数据质量。项目研究随机介质中成像机理，探索图像退化规律，并恢复图像，使其达到或近似达到衍射极限分辨率。首先，研究随机介质中共轭关系退化机理，描述随机介质导致的物-像之间几何关系变化的规律。其次，研究随机介质中辐射能量分布退化机理，提出域稳定约束条件，估计退化函数支持域尺寸和参数，表现图像亮度分布退化的统计规律。最后，在这两个模型基础上研究恢复观测图像的解相关方法。本项目研究是图像序列处理研究和应用的理论基础的重要组成部分。研究成果有望给计算机视觉和自适应信号处理研究带来新发展。

在远距离成像过程中，图像序列受到随机介质因素，如大气湍流、雾、沙尘、浑浊水体等的影响会出现像素点亮度的随机起伏、闪烁和图像中物体的位置漂移，这使得传统的背景建模方法在随机介质环境下难以准确突出目标的视觉表现，给目标检测、跟踪和识别带来困难。针对图像受到随机介质影响在不同区域表现出的不同性质，项目研究提出分层次决策判别方法，用高斯模型建模背景平坦区域，用双高斯模型建模背景中的物体边缘区域，设定判别式对每个像素点进行判别，并在线更新模型参数。针对由于随机介质影响出现的亮度突变点建立自适应判别模型，结合前一层次的判别结果，构造判别条件消除亮度突变点，提取目标点。最后通过连通区域约束得到目标区域，恢复目标区域的图像，项目研究得到的方法在不同随机介质条件都取得了良好的图像恢复效果。